I. INTRODUCTION

During recent decades, interest in the study of household finance has grown rapidly. Campbell (2006) first advanced the case for treating household finance as a distinct field of study in economics. The global financial crisis of 2008-2009 strengthened that case due to the subprime housing debacle in many industrial economies and its persistent impact on household balance sheets. In particular, the extent and nature of increased leverage and risk in household mortgages and their effects on the real (housing industry) and financial (shadow banking) sectors of the economy were not well known or understood prior to the crisis. Consequently, there is now a focus on household decision making, how households got into this trouble, what transpired in the crisis, and the difficulties encountered thereafter. (1)

A hindrance to research and understanding of household economic behavior (real and financial) has been the lack of sufficient data. Relative to other countries, the United States has a large amount of high-quality data on household economic behavior; these data will be examined closely in this paper. Even the U.S. data, however, were inadequate to inform economic agents and policymakers sufficiently to avoid the financial crisis. Many efforts are underway to acquire and develop additional needed data; these efforts include the Eurosystem's Household Finance and Consumption Survey (HFCS), which was inspired partly by the U.S. Survey of Consumer Finances (SCF). (2) Other efforts, such as the National Academy of Science's call for a substantially revised Consumer Expenditure Survey (CE), aim to reform existing datasets (Dillman and House 2013).

The U.S. household survey data exhibit several characteristics that limit their effectiveness. The U.S. statistical system (public and private) is decentralized, with each data source specializing in a part of household activity. Although there are often good reasons for specialization, the result is a general lack of comprehensive measurement of household activity. Many datasets are cross-sectional, which limits their ability to track the behavior of specific households over time, and are gathered infrequently. When data sources are combined in an effort to provide a more comprehensive view of household behavior, the combination of the specialized data sources can create imperfect, if not misleading, views of household economic conditions, due to differences in sampling, measurement, and linkages between microeconomic and aggregate data. (3) These imperfections make it difficult to ascertain from the data the extent and nature of important developments, such as adjustments affecting household balance sheets in the wake of financial crisis, increases in income inequality, and intergenerational dynamics of household net worth.

Data on household behavior in other countries also exhibit limitations, but there are signs of improvement in response to major economic developments. Most notably, the financial crisis reaffirmed the view that household finance is at the center of development economics because financial access is thought to be one of the key factors that could help poor and vulnerable households become more productive and resilient in the face of economic shocks. In addition, there have been payment innovations such as M-Pesa in Kenya, an electronic money issued by a cell phone company, Safaricom, that in many respects is now on par with currency there as a medium of exchange (Jack, Suri, and Townsend 2010). The often-expressed hope in developing economies is that a deeper, more developed financial system can be built on top of such an improved payments system, with some progress evident in countries such as Pakistan. (4) These developments bring us back to the need for better data on payments, household behavior, and a microfounded view of the macroeconomy in developing countries. Fortunately, more countries are producing data from household surveys that are doing a better job of measuring these developments.

We believe an important step forward in understanding household behavior is the development of more reliable and effective measures of household economic activity, both real and financial. Therefore, an overarching goal of this paper is to describe a comprehensive vision for practical implementation of household surveys that are integrated with financial statements and payments data, leaving no gaps in measurement and strengthening the theoretical and applied linkages among measures. The main contributions of this paper are: (1) to assess how well integrated U.S. household surveys are with elements of financial statements for households: and (2) to demonstrate how a diary of U.S. consumer payment choices can be used to construct a new statement of liquidity flows that advances the current state of the art in measuring stock-flow dynamics and thus takes a step closer to realizing the overarching vision of the paper.

Samphantharak and Townsend (2010, henceforth ST) describes the baseline conceptual framework for the design of an integrated survey that has been implemented in the field for almost 20 years and that allows construction of a complete set of household financial statements that is comprehensive and fully integrated. Essentially, ST create a set of financial accounts akin to those of corporate firms: this set comprises a balance sheet, income statement, and statement of cash flows (CF). The concept is of a household with projects, that is, a collection of assets that earn income from farm and nonfarm production activities. This idea of assets earning income also extends to households engaged in wage or salaried labor, meaning those that essentially generate income from their human capital. A key element of this analysis is that all aspects of household situations and behaviors are measured: income, in order to measure the productivity of physical and human capital; assets and liabilities, to measure wealth; and CF, to distinguish liquidity from income and profitability. A key to this measurement is that the accounts are required, by construction, to be consistent with one another, thereby eliminating the possibility of gaps. Few surveys feature this dynamic integration.

To illustrate how this works, and as a first step in the paper, we use the ST framework to assess the degree of integration in leading U.S. household surveys. For each survey considered, we tabulate and juxtapose the data of each in the form of corporate financial statements applied to the representative U.S. household. We first construct for each survey a harmonized balance sheet, income statement, and statement of CF for a recent time period that matches the survey dates--around 2012--as closely as possible. To ensure maximum accuracy, we have invited assistance from representatives associated with each survey; and to encourage further refinement of this effort, we make our programs available to interested researchers. Then, we use the estimated U.S. household financial statements to characterize the degree of integration by two distinct measures. Integration by coverage reflects the extent to which a survey contains estimates of each line item in the financial statements. All the surveys cover roughly half the income statement items, although most specialize in income or expenditures. However, the coverage of the balance-sheet items varies widely across surveys. Integration by dynamics reflects the extent to which the statement of CF accurately measures the law of motion between stocks (shown in the balance sheet) and flows (shown in the income statement). None of the surveys can provide truly direct statements of CF, and all of them make large errors relative to indirect estimates of changes in assets and liabilities.

Our assessment of integration in U.S. household surveys is merely a factual statement of results and is not intended to be a criticism of the surveys or a call for reforming them. We recognize and accept the specialty nature of U.S. surveys, which has the benefit of allowing gains from specialization and achievement of each survey's original goals. For example, the Panel Study on Income Dynamics (PSID) was originally designed to measure poverty and to contribute to its reduction in conjunction with President Johnson's Great Society programs; the CE was designed to gather data for developing accurate price indices; and the SCF to measure wealth. Although some of these surveys have evolved over the years, particularly the PSID, others retain their original mandate. Yet, the specialization and persistence of the U.S. surveys do leave gaps in measurement that can only be overcome by comprehensive integration of the surveys with financial statements. Ironically, because the PSID and SCF are so highly regarded, they are adopted as the gold standard elsewhere in the world, for example, in China and Europe, thus propagating essentially the same gaps in these other surveys as in their U.S. counterparts.

A second step of this paper is to use the Federal Reserve Bank of Boston's 2012 Diary of Consumer Payment Choice (DCPC) to demonstrate how consumer payment diary surveys can improve the dynamic integration of surveys. (5) The DCPC directly measures several, but not all, components of the law of motion governing the stock-flow relationship between assets and liabilities (balance-sheet items) and income and expenditures (income-statement items). Because the 2012 DCPC is focused on consumer payments authorized by payment instruments (cash, check, debit or credit card, online banking, and such), it focuses on liquid assets used as payment instruments, including the currency held and used by U.S. consumers. In this respect, the DCPC is similar to the Townsend Thai Monthly Survey (TTMS), which underlies the ST methodology, where currency is the main household asset and payment instrument in rural Thailand. To provide a bridge to our key next step, we compare and contrast the household financial statements constructed with TTMS with those constructed with the DCPC.

The central innovation of this paper is the construction of a new, more detailed analysis of CF at the level of liquid asset accounts, where currency, checking accounts, and other liquid assets are distinguished and treated separately. By tracking consumer expenditures that are authorized by payment instruments tied to specific types of liquid asset accounts, the DCPC matches expenditures to the sources of money and credit that fund them. This matching cannot be done feasibly by surveys that track consumer expenditures at the level of individual products (the CE) or at the level of aggregated expenditure categories ("food away from home").

Linking all the liquidity accounts to one another and to the expenditures (or investments) they fund makes it possible to better assess the changing landscape of payments taking place in the United States and industrialized countries as well as in emerging-market and low-income countries. (6) This then links back to the need for data to better inform public policy and to provide consumers with the information they need to improve household decision making and economic behavior. More informative financial accounts come from considering payments, and vice versa: better payments data come from integrated financial accounts. Development of household economic data from dynamically integrated household surveys that include payment diaries might be particularly beneficial for developing countries, where household economic data are scarce, there are few preestablished surveys with prior missions, and payment systems and financial industries are changing rapidly. Of course, payments systems are also changing in the United States. The 2015 DCPC took a small step toward integrating payments and employing the ST framework, as described below. We provide a framework and guidance for policymakers to implement this longer-run vision.

The remainder of the paper proceeds as follows. Section II provides an overview of the main U.S. household surveys. Section III reviews the ST methodology and explains how it will be used in our analyses. Section IV assesses the degrees of integration in U.S. household surveys, by coverage and dynamics. Section V compares and contrasts the TTMS and DCPC survey data. Section VI describes the innovation made possible by the interaction of ST's methods with the DCPC. Section VII concludes.

II. OVERVIEW OF U.S. HOUSEHOLD SURVEYS

This section describes the main surveys included in this study, which are used to collect data on U.S. household economic conditions (henceforth, "household surveys"), plus the TTMS. Summary descriptions of these surveys appear in Table 1 in order of chronology based on continuous fielding. Five sponsors produce these U.S. surveys:

* University of Michigan, Institute for Social Research (ISIR)--The Michigan ISIR sponsors two surveys. First, the biennial PSID, which is "the longest running longitudinal household survey in the world" and that includes data on wealth and expenditures as well as other socioeconomic and health factors. (7) Second, the biennial (even-numbered years) Health and Retirement Survey (HRS), which "has been a leading source for information on the health and well-being of adults over age 50 in the United States" for more than 20 years; the HRS includes the biennial Consumption and Activities Mail Survey (CAMS) for tracking household expenditures in "off' years (odd-numbered). (8)

* U.S. Bureau of Labor Statistics (BLS)--The BLS sponsors the CE, comprising "two surveys--the quarterly Interview Survey and the Diary Survey--that provide information on the buying habits of American consumers, including data on their expenditures, income, and consumer unit (families and single consumers) characteristics." (9) "As in the past, the regular revision of the Consumer Price Index (CPI) remains a primary reason for undertaking the Bureau's extensive CE. Results of the CE are used to select new 'market baskets' of goods and services for the index, to determine the relative importance of components, and to derive cost weights for the market baskets."

* Federal Reserve Board--The Board sponsors the SCF, "normally a triennial cross-sectional survey of U.S. families. The survey data include information on families' balance sheets, pensions, income, and demographic characteristics. Information is also included from related surveys of pension providers and the earlier such surveys conducted by the Federal Reserve Board." The SCF collects some consumer expenditures directly. (10)

* U.S. Census Bureau--The Census Bureau sponsors the Survey of Income and Program Participation (SIPP), "the premier source of information for income and program participation. SIPP collects data and measures change for many topics including: economic well-being, family dynamics, education, assets, health insurance, childcare, and food security." (11)

* Federal Reserve Bank of Boston--The Boston Fed's Consumer Payments Research Center (CPRC) sponsors the annual Survey of Consumer Payment Choice (SCPC) and the occasional DCPC, both of which measure consumer adoption of payment instruments and deposit accounts and the use of instruments. Originally, the SCPC and DCPC were not integrated like the CE but were developed independently; they are now being integrated. The SCPC collects only the number of payments, while the DCPC also tracks the dollar values. Both provide data on cash and (in later years) checking accounts plus revolving credit. The SCPC contains very limited information about household balance sheets.

These surveys were selected because of their quality and breadth of coverage of U.S. household financial conditions, including relatively large numbers of detailed questions pertaining to the line items of household financial statements (assets, liabilities, income, or expenditures). None of the surveys contains all relevant financial conditions because none was designed to do so. Thus, no single survey is fully integrated with financial accounting statements and no single survey alone can provide complete estimates of household financial conditions. When combined, however, these U.S. household estimates come closer than any single dataset available today to providing a comprehensive assessment of U.S. household financial conditions. These surveys were also chosen because, except for the HRS, they are representative of U.S. consumers. (12) However, the surveys are implemented with different samples of households (or consumers) and, in some instances, substantively different survey questions, so their estimates are not necessarily comparable.

We reiterate that each survey has its own particular purposes or goals and that none is intended to provide a comprehensive, integrated set of household financial conditions as described in ST. The CE, for example, is primarily intended to produce data on a wide range of consumption expenditures that aid in the construction of the CPI. In contrast, the SCF primarily tracks details of assets and liabilities plus income from all sources but does not track all consumer expenditures. The PSID aims to estimate most income and expenditures but also focuses on collecting data on social factors and health, a practice that might be beneficial for every survey and data source. In any case, the PSID's breadth limits the amount of detail it can obtain on income and expenditures, so it does not obtain a comprehensive estimate of balance-sheet items. For all of these reasons, the analysis in the next section does not expect or presume to find an individual integrated financial survey, nor does it recommend that any of these surveys change what it is currently doing.

Table 1 summarizes the key characteristics of the selected U.S. household surveys in terms of their basic features, survey methodologies, and sampling methodologies. Surveys are listed in columns in chronological order (left-to-right) based on their initial years of continuous production. The oldest is the PSID, which dates back to the 1960s, while the newest, the SCPC and DCPC, are less than a decade old. Most of the surveys are conducted relatively infrequently, ranging from quarterly (the CE and SIPP) to triennially (the SCF). Although implemented daily for 1 or 2 months, the official DCPC has been implemented only three times in 5 years. The date of statistical calculations refers to the period used to estimate the elements of the household financial statements, as discussed later in the paper. The rows of the table are grouped into sections related to the survey methodology and the sampling methodology. For further comparison, the table also shows corresponding information about the TTMS.

Survey methodologies vary widely across the surveys along several dimensions. One obvious distinction is the mode: survey (PSID, CE-S, SCF, HRS, SIPP, and SCPC) versus diary (CE-D and DCPC) or "diary survey." This distinction is complicated by the fact that modes also vary for each type of survey or diary, including paper surveys, paper diaries (or memory aids), online surveys--with or without assistance--and interviews; some surveys use mixed-mode strategies. A key differentiating factor among surveys is whether they collect data based on respondents' recall, where the recall period can vary in length from a period of 1 week to 1 year, or based on respondents' recording the data, where the recording period is typically 1 day. Recall-based surveys are more susceptible to memory errors and aggregation errors (over time and variable types). Some sponsors field their own survey (Michigan ISIR), while others outsource to vendors (e.g., the SCF uses NORC, formerly called the National Opinion Research Center).

The sampling methodologies are relatively similar across surveys. All surveys aim to provide estimates that are representative of some U.S. population measure, except the HRS, which is limited to older households. The main reporting unit varies across surveys from individual consumers to entire households, with some surveys obtaining information about the household from just one member--an important choice that can significantly affect the results of the survey. The surveys also differ in whether the samples are drawn as independent cross sections or as longitudinal panels. The precision of survey estimates varies widely because sample sizes range from 2,000 to 52,000 reporting units.

Estimates of economic and financial activity for consumers and households are influenced heavily by at least two major types of factors: (1) heterogeneity in the survey specifications, sampling methodologies, and data collection methodologies; and (2) variation across surveys in the content, scope, and nature of questions about real and financial economic activity. Therefore, the reader should not expect estimates of income, expenditures, or wealth from the surveys to coincide. Instead, there might be large discrepancies in estimates of these economic and financial activities even if the conceptual measures are similar. Differences in target populations can naturally produce large differences in economic and financial measures. But even more subtle survey design differences, such as recall versus recording, can produce large differences in the estimated measures. With regard to survey content and questions, even minor differences in wording can elicit differences in measured concepts between surveys. Similarly, the level of aggregation--collecting data on just the total or on the sum of the parts of the total (and then adding them up)--can have dramatic effects on estimates of the total values across surveys.

III. THE ST FRAMEWORK

This section provides a brief overview of the ST (2010) framework for defining and measuring the integration of household surveys with corporate financial statements.

A. Conceptual Framework

There are three main financial statements in the ST "household as corporate finance" framework. (13) The first statement is the balance sheet or the statement of financial position, which reports all assets and liabilities at a point in time. The difference between assets and liabilities is net worth. In the terminology of corporate financial accounts, net worth is the household's equity in the household enterprise. The second financial statement is the income statement, which measures flows of revenues and expenses as well as the disposition of net profit into consumption and savings over a period of time. Finally, the statement of CF measures money, cash, or other liquid assets flowing into and out of the household as part of the payments system. In practice, CF are simply the outflows of cash for the acquisition of inputs of production, as well as for investment and consumption expenditures, and the inflows from sales of product, liquidation of assets, and financing.

The balance sheet is a stock report, while the income statement and the statement of CF are flow reports. There is a close connection between the balance sheet and the income statement through the connection between stocks and flows, as summarized in Figure 1. Specifically, profits from production or from salary and other income can be saved or consumed. Consumption is analogous to paying out a dividend to the owner. Positive savings show up as an increase in (real or financial) assets and wealth, reflected in the balance sheet at the end of the period. Likewise, negative savings show up as a decrease in assets and wealth. Indeed, the change in wealth in the balance sheet between two points in time is essentially net savings. (14)

Income in corporate financial statements is typically accrued income, based on the idea that expenses of production are not subtracted until revenues from sales resulting from that production are recognized. (15) The essential idea behind this notion of accrued income is that one wants to measure the ultimate return on a project in order to compare that return to alternatives; that is, one wants to measure the opportunity cost in order to see whether the project is warranted, in order to answer the obvious question: do the economic activities the household has adopted "make sense"? Essentially, accrued income is supposed to measure productivity. However, since the accrual basis of accounting does not necessarily recognize revenues or expenses when cash flows in or out of the enterprise, it cannot give analysts a full understanding of the enterprise's liquidity. For example, a project may be productive with a reasonably high rate of return, but it may become illiquid due to CF fluctuations and the household may even go bankrupt. This example illustrates one of the reasons why the statement of CF is needed to obtain a comprehensive understanding.

To summarize, the reconciled financial statements must exhibit the following accounting identities: (1) in the balance sheet, the household's total assets must be identical to its total liabilities plus total wealth or net worth, (2) the increase in household wealth in the balance sheet over the period must be identical to the household's savings (adjusted for unilateral transfers); that is, it must be identical to a household's net income from the income statement minus consumption, and (3) the increase in the household's cash holdings in the balance sheet must be identical to the household's net cash inflow in the statement of CF, summing over all sources. Both sides of every accounting identity are measured.

One benefit of imposing accounting identities is that we avoid the common problem that a variable generated from one set of questionnaire responses yields a different value when computed from an alternative set of responses. For example, Kochar (2000) finds that household savings in the Living Standard and Measurement Study (LSMS) surveys computed as "household income minus consumption" is different from household savings computed from "change in household assets." This discrepancy could come from various problems in questionnaire design. For example, some of the assets might be omitted from total assets, some assets might be financed by liabilities rather than savings, or income and savings might be defined inconsistently. Indeed, as mentioned above, one can use these two different measures of savings, which may differ as indicated, as a consistency check within a survey or diary fielding, with follow-up questions in the case of discrepancies.

ST applied this vision of integrated surveys to the TTMS. Transactions in the monthly data are like journal entries for an accountant, allowing the analyst to create complete financial accounts. As details of the transaction partners are also recorded, one can map networks within the village and also geographic patterns. Figure 2 illustrates the procedure for creating a household's balance sheet, income statement, and statement of CF from a panel household survey. More information about the TTMS appears in Section V.

B. Details of the Statement of CF

Because the dynamic accounting of linkages between stocks and flows is central to this paper, we provide a more detailed discussion of this topic. The statement of CF provides an accounting of cash received and cash paid during a particular period of time, thereby providing an assessment of the operating, financing, and investing activities of the firm (or household).

The first step in constructing a CF statement is to define the term "cash." Despite the label, it is important to remember from the outset that currency is typically only part of this. For advanced industrial economies such as the United States, standard corporate financial statements tend to focus CF on the concept of "cash and cash equivalents" (CCE):

* Cash--Currency (coins, notes, and bills) (16) and liquid deposits at banks and other financial institutions, including demand deposits, other checkable deposits, and savings accounts. This measure is similar to the broad measure of money known as M2. (17)

* Cash equivalents--Short-term investments with a maturity of 3 months or less that can be converted into cash quickly, easily, and inexpensively (high liquidity and low risk). None of the surveys identify cash equivalents separately from similar investments of longer maturity. Examples include 3-month Treasury bills versus 1-year Treasury bonds and 3-month versus 6-month certificates of deposit. (18)

The assessment of U.S. surveys will focus on CCE for the statement of CF. For the TTMS and DCPC, however, the statement of CF will focus only on currency because Thai households transact primarily in currency (Thai baht) and the 2012 DCPC is a payment diary that does not track the entire balance sheet and has only one liquid asset (currency in U.S. dollars, which is a payment instrument). (19) Most U.S. surveys do not collect data on currency, which is a relatively small portion of liquidity for most U.S. households, and only the SCPC and DCPC do so comprehensively.

Once cash is defined, CF for that defined concept (CCE) can be calculated to account for the operating, investing, and financing activities of the firm (or household). (20) In particular, the statement of CF includes three main parts:

* CF from production (or operating activities)--These are net CF from operating activities of the firm (or household). The direct method shows cash inflows from operations and cash payments for expenses, by major classes of revenue and expense. Equivalently, the indirect method converts net income from an accrual basis to a cash basis, using changes in balance-sheet items.

* CF from investing activities (consumption and investment)--These are net CF from investing activities of the firm (or household). Cash outflows are primarily for investment in capital and for the purchase of securities that are not CCE. Cash inflows are the converse, including sales of capital and non-CCE securities. Individual items are listed in gross amounts (inflows minus outflows), by activity. As applied to the household, these are consumption expenditures (on nondurable goods and services) and capital expenditures (on durable goods).

* CF from financing--These are net CF from transactions considered to be the financing activity of the firm (or household). Cash inflows occur when resources are obtained from owners or investors, such as by issuance of equity or debt securities. Cash outflows are the converse, in the form of payment to owners and investors or to creditors. As with CF from investing, individual items are listed in gross amounts.

Another type of transaction sometimes associated with the statement of CF is direct exchange, which occurs when noncash (not CCE) assets or liabilities are traded without implications for cash. Often these exchanges are difficult to classify as either investing or financing activity because they may have elements of both. For that reason, accountants do not agree on whether to include direct exchanges in the statement of CF or to report them in a separate statement. For this paper, we do not include them in the statement of CF.

In theory, the statement of CF provides an exact linkage between flows in the income statement and changes in stocks on the balance sheet. To verify this, the statement of CF compares measured CF with the measured changes in assets and liabilities from the balance sheet. Total CF is simply the sum of component flows,

[CF.sub.t] = [CF.sup.p.sub.t] + [CF.sup.v.sub.t] + [CF.sup.f.sub.t],

where superscript p denotes production (operating activity), v denotes investing activity, and f denotes financing activity. If all financial-statement items are measured accurately and constructed comprehensively, this estimate from the statement of CF should exactly match the change in the stock of cash from the balance sheet,

[CF.sub.t] = [DELTA][A.sup.C.sub.t] = [A.sup.C.sub.t] - [A.sup.C.sub.t-1],

where [A.sup.C.sub.t] denotes the asset value (end-of-period t) of CCE (superscript C). If these CF identities were to hold exactly using data from a survey, then that survey would be fully dynamically integrated with financial statements. In practice, however, measurement of financial-statement items is neither exact (due to measurement error) nor comprehensive in actual surveys (due to failure to include all items), so we expect to observe errors in the CF identities above (i.e., we expect to see less-than-full dynamic integration). One logical measure of the degree to which survey estimates are integrated across time (dynamically) is

CF error = 100 x [[CF.sub.t] - [DELTA][A.sup.C.sub.t]/[A.sup.C.sub.t-1]]

which is expressed as a percentage of lagged cash. Smaller CF errors (in absolute value) are interpreted as indicating better dynamic integration of a survey. (21)

This analytical linkage between CF (also on the income statement if the cash basis rather than the accrual basis is used) and the stock of cash (balance-sheet items) can be disaggregated into the linkages between individual liquid assets (stocks) in CCE and the gross flows among them. Henceforth, our language assumes the cash basis is used, but our analysis remains valid for the accrual basis, since the real difference between the cash and accrual bases is only the labeling of the transaction; for example, goods sold create an account receivable that is not necessarily cash and does not appear on the statement of CF if the latter does not recognize accounts receivable as CCE. Nevertheless, the sale would be recognized as creating an increase in an asset (an accounts receivable item).

To see the point about disaggregation, let [A.sup.C.sub.kt] denote the end-of-period dollar value of a liquid asset in CCE from the balance sheet, where subscript k denotes the account/type of liquid asset (currency, demand deposits, and such) and subscript t denotes the discrete time period (such as month, quarter, or year). Liabilities, [L.sub.kt], are defined analogously and primarily represent various types of loans; in principle, liabilities can be viewed as negative-valued assets. (22)

Let [D.sub.kdt] denote the dollar value of deposits into account k on day d (nearly continuous), and Wkdl the analogous withdrawals. (23) Gross CF in period t are the sums across all daily flows into and out of an asset type:

[mathematical expression not reproducible]

Asset deposits include primarily income of all types (including any capital gains and losses from holding CCE), transfers of another type of asset (or liability) into the account, or unilateral gifts received. Asset withdrawals include primarily payments for goods and services (consumption expenditures or capital goods investment), transfers to another type of asset, or unilateral gifts given. Again, liability flows are defined analogously.

Individual assets are governed by the following law of motion between periods t - 1 and t:

[A.sup.C.sub.kt] = [A.sup.C.sub.k,t-1] + [D.sub.kt] - [W.sub.kt] [DELTA][A.sup.C.sub.kt] = [D.sub.kt] - [W.sub.kt].

Individual liabilities are governed by an analogous law of motion where the liability "return" is primarily interest paid.

Finally, the disaggregated CF for each CCE type of asset include some that net to zero when aggregated across all account k accounts. For example, if a consumer withdraws $100 in currency (k = 1) from a checking account (k = 2), then [D.sub.1dt] = [W.sub.2dt]. For this reason, it is informative to track the flows among types of asset (and liability) accounts when analyzing the CF behavior of households. For some types of asset accounts, such as a checking account, withdrawals can be made with multiple payment instruments, such as checks, debit cards, and various electronic bank account payments. Thus, the gross flows between accounts can be further disaggregated by the type of payment instrument used to authorize the flow. (24)

IV. ASSESSMENT OF INTEGRATION IN U.S. HOUSEHOLD SURVEYS

This section evaluates the content and structure of the main U.S. household surveys, excluding the SCPC and DCPC, which are not designed to be general surveys of household finance, in relation to corporate financial statements. As noted earlier, no U.S. survey is fully integrated with financial statements in a manner consistent with the ST framework. However, all of the U.S. surveys contain questions that provide estimates of many of the relevant stocks and flows in financial statements. Therefore, the ST framework can be used to organize the survey data into estimates of a representative (average) U.S. household's financial statements: a balance sheet, income statement, and statement of CF. The remainder of this section presents those estimates for each survey and analyzes the results.

The tables in this section report estimates of U.S. financial statements from the surveys. Each statement contains nominal dollar-value estimates for the line-item elements from each survey, aggregated to the U.S. average per household, with the sampling weights provided by the survey programs. (25) Selected aggregate measures are supplemented with medians. The line items (rows) of each financial statement reflect our best effort to combine survey concepts into reasonably homogeneous measures. (26) Where necessary and feasible, some survey concepts fall into the "other" categories; tables are footnoted extensively to clarify these details. To the extent possible, all economic concepts from each survey are included in the statements. However, the question wording and concept definitions can vary significantly across surveys, so detailed estimates fall short of perfect harmonization. To ensure proper handling, we have provided our preliminary results and software programs to managers or principal investigators of each survey and offered them the opportunity to evaluate and correct our analysis. (27)

Juxtaposing estimates of the financial statements for each survey provides two benefits. First, and independently of the ST methodology, the financial statements provide valuable information about the relative magnitudes of real and financial economic conditions estimated by each survey. Differences between survey estimates can be large in absolute and relative terms because of the absence of perfect harmonization, as noted above. The aggregate estimates may also diverge due to significant differences in survey or sampling methodologies, described in Section II, or due to differences in the coverage of statement line items, described below. In any case, the comparison of estimates reveals the relative strengths and weaknesses of each survey in measuring household economic conditions.

Second, juxtaposing the estimates facilitates an easy and quantitative assessment of how well each survey's questions integrate with the elements of the household financial statements. The degree of integration can be evaluated by at least two standards: (1) the coverage of items in the statements; and (2) the dynamic interaction between stock and flow concepts. With regard to coverage, we can further quantify two types of coverage: (1) the percentage of detailed line items estimated by the survey; and (2) the aggregate dollar values of the estimates. As an example of the first of these coverage measures, suppose that a balance-sheet concept had ten detailed items and one survey estimated eight of them while another estimated only two of them. Then, the first survey has broader coverage (80% vs. 20%). However, line-item coverage is not necessarily an accurate indicator of value coverage. If a survey had two estimates of the ten balance-sheet items, and if each one were an estimate of the aggregate of five of the detailed items (e.g., short-term assets and long-term assets), then the survey might produce a very high percentage of the total value of assets even though it did not include an estimate of each of the ten items. Still, estimating the aggregate value of five items without estimating each individual item is prone to producing biased estimates due to the adverse effects of recall and reporting errors. The juxtaposed estimates reveal the extent to which this kind of aggregation effect appears in the survey estimates.

A. Balance Sheets and Income Statements

Balance sheets constructed from the U.S. surveys appear in Table 2 (A [assets] and B [liabilities]). The asset and liability estimates are reported as current market values to the best of our ability, although it is not always possible to be certain of the type of valuation reported by respondents. Assets are divided into financial and nonfinancial categories, with financial assets further divided into highly liquid current assets (short-term) and assets with other terms and liquidity (long-term). For financial assets, surveys usually obtain market values explicitly or by assumption; where they distinguish between face value and market value (e.g., for a U.S. government saving bond) the latter is reported. For nonfinancial assets, the valuation issue is almost the same, except the potential distinction is between market value and book value. (28) For housing assets, the surveys generally ask for the current (market) value of homes, but we cannot be sure they do not report the purchase price, which is a book value. For business assets, all surveys ask for a current (market) value, although the form of the question varies and may use analogous terms (e.g., "sale price"). Liabilities are the current outstanding balances for debt, not the original loan amounts. Liabilities are divided into categories of revolving debt, characterized by an indefinite option to roll over the liability, and nonrevolving debt. Because the maturity of debt is generally not known from the surveys and the term varies by debt contract within a category, the nonhousing debt categories are listed in rough order of liquidity from most to least liquid.

All the surveys report an estimate of total assets in Table 2A. The U.S. households own average assets worth as much as $632,246, according to the SCF, less half that amount, $226,314, in the CE survey. The HRS estimate of $556, 295 is close to the SCF estimate, despite being limited to older consumers. The breakdown of asset types is similar for all the surveys. Financial assets generally account for less than half of asset values, 29%-41%, despite variation in the number and type of detailed asset categories. Tangible (physical) assets represent the majority of asset values. Within financial assets, cash accounts for roughly $30,000 for all but the SIPP, where it accounts for roughly $12,000, and most is held in bank accounts. Only the SCF contains an estimate of currency, but even that is not a direct estimate of actual currency holdings of the household. (29) Overall, estimates of balance-sheet assets are relatively comprehensive for all surveys, as shown by their similar aggregate values and by the breadth of coverage across detailed asset categories. The SCF is the most comprehensive, with asset estimates in every category except short-term assets other than bank accounts (checking and saving); the PSID, HRS, and SIPP are almost as comprehensive as the SCF. The CE is much less comprehensive and has considerably lower asset values.

All the surveys also report an estimate of total liabilities. The U.S. households have average liabilities ranging across the surveys between $61,979 and $ 112, 306, much lower than the value of total assets and exhibiting less variation than across surveys. Housing debt is by far the largest portion of liabilities, ranging from $58,143 to $87,228 in all surveys where it is reported. The HRS asks specifically only about housing-related debt, with a catch-all question for other loans. The SIPP does not permit an exact estimate for housing-related debt, but the "other loans" category most likely includes some housing-related debt. While estimates of balance-sheet liabilities are somewhat comprehensive for most surveys, they are not as comprehensive as the estimates of assets. The aggregate values vary less and there is less line-item coverage across detailed categories of liabilities. Once again, the SCF is the most comprehensive, with liability estimates in nearly every category. The PSID is almost as comprehensive as the SCF. The other surveys are less comprehensive, although in different ways. Given the estimates of total assets and total liabilities, household net worth ranges from $152,646 in the CE to $519,940 in the SCF.

Income statements constructed from the U.S. surveys appear in Table 3. Income is divided into two main categories: compensation of employees (the most common source of U.S. household income) and other income. The latter includes income from all types of businesses owned and operated by households. Expenditures also are divided into two main categories: production costs and taxes. As explained above, the production costs of households are expenditures associated with businesses operated directly by a U.S. household; these businesses include sole proprietorships, partnerships, and certain Limited Liability Corporations (LLC). (30) Unlike TTMS, where most households operate a business (typically agricultural), only a minority of U.S. households have a business. (31) For the minority of U.S. households with a business, it would be natural to apply corporate financial accounting to income (revenues) and expenses, as in ST. However, none of the surveys provide sufficient information about household business activity, so we use the simpler approximation of revenues as "income" to accommodate the majority of U.S. households without a business. Furthermore, all income-statement estimates are reported on a cash basis of accounting, so revenues and expenses are reported for the period when the cash is received (income) or paid out (expenditures), because this method is the primary way data are collected in the U.S. surveys.

All of the surveys report an estimate of total income (revenues). The U.S. households received average total income of $61,431 to $83,863 per year. Estimates of labor income are even more similar across surveys, ranging only between $42,377 and $53,623, essentially all of which is wages and salaries. Estimates of other income types vary more, ranging between $9,816 and $37,402, but account for less than one-quarter of total income, except for the HRS estimates, which represent 45% of total income. Overall, income estimates are the most comprehensive and consistent portion of the household financial statements across surveys, most likely because employment compensation is widespread among U.S. households and the data are relatively easy to collect. Estimates of income other than employment compensation are less uniform across the surveys due to the unavailability of some detailed line-item categories.

Although three surveys (the PSID, CES, and SCF) have estimates of business income, none of them provide much information about household business expenditures. They ask few, if any, questions about household business activity (aside from the mere existence of a home business). No survey has an estimate of production costs for household businesses. Only three surveys with business income have estimates of taxes (these estimates average less than $5,000 per household), and only the CE reports employment taxes. Tax expenditures are those paid directly by households and do not include taxes deducted by employers or paid by third parties on behalf of households.

Given their estimates of total income and total expenditures, all of the surveys provide estimates of net income (income less expenditures), which range from $60,971 (CE) to $81,856 (SCF), as shown at the bottom of Table 3. The HRS does not collect expenses, so its net income equals total income. Net income is similar to income in the other surveys because expenditures are relatively small (taxes only). Household net income is treated as retained earnings that are distributed to household members for consumption and investment expenditures, which are recorded in the statement of CF (described below).

B. Quantifying Integration by Coverage

We wish to characterize the degree to which surveys are integrated with household financial statements in terms of coverage. We propose to develop the criteria for measuring this kind of integration by quantifying the extent to which a particular household financial survey covers (includes) the breadth of the line items in standard balance sheets and income statements. There are at least two dimensions along which integration by item coverage could be measured using the estimates from the preceding subsection. One is the fraction of detailed line items for which a survey provides estimates ("line-item coverage"). Another is the fraction of the total dollar value of all line items estimated by a survey ("value coverage"). The two measures are independent and not necessarily highly correlated. A survey could cover most items in the financial statements but underestimate them significantly; likewise, a survey might cover only a small number of items but obtain very high-value estimates if the items covered include mainly the highest-valued items. The latter situation may occur when a survey only collects data on two aggregate subcategories (such as short-term and long-term assets) but collects none on the detailed line items within each subcategory.

We construct the measure of line-item coverage as follows. We define the range of each financial statement as the number of the most detailed line items (rows) from the tables earlier in this section. Then, we count the number of line items (rows) for which each survey provides a dollar-value estimate. The coverage estimate of integration is the proportion of line items estimated relative to the total number of line items. We call this the "item-coverage ratio," and we construct two separate ratios, one for the balance sheet and one for the income statement. This measure reflects only the extensive margin of coverage because it does not account for the magnitude of the dollar values in each line item; thus, it may not give a complete reflection of coverage for total assets, liabilities, income, or expenditures.

We construct the measure of value coverage analogously, as follows. We use the nominal dollar values for each individual line item in the statements to construct the aggregate total values (sum of all individual items) for each statement and divide the aggregate value by the best available per-household estimate of the relevant metric for the U.S. population. For the balance sheet, we use total assets and total liabilities from the Flow of Funds accounts as the denominator. For the income statement, we use personal income from the National Income and Product Accounts (NIPA). The "value-coverage ratio" represents survey coverage of the intensive margin of coverage. The difference between the two types of ratios reflects the extent to which a survey's coverage of financial statements is more integrated in its intensive or extensive coverage of financial statements. To the extent that one wishes to construct accurate estimates of aggregate U.S. household financial conditions, the dollar-value ratio may be more important.

Figure 3 provides scatter plots of the item-coverage ratio (diamonds) and value-coverage ratio (squares) for the balance sheet and income statement. The feasible range of both ratios is [0, 1], with the upper end indicating that a survey has estimates of every single item in the corresponding financial statement. Recall that the ratios are independent and may not be highly correlated. Thus, the item-coverage ratio does not necessarily reflect how well a survey produces aggregate estimates of the data, and the value-coverage ratio does not necessarily reflect how well a survey covers the number of line items in the financial statements. Also, we make one important adjustment to the income statement ratios to adjust for the application to households. As shown in the next subsection, household consumption and durable goods investment are listed in the statement of CF rather than the income statement. However, for the purpose of quantifying the overall coverage of household income and total household expenditures, both business-related expenditures and household consumption or investment expenditures, we include all types of expenditures in constructing the coverage ratios for the income statements.

None of the U.S. surveys are completely integrated (ratio of 1.0) with aggregate financial conditions for either statement, as can be seen from Figure 3. In fact, no survey has either type of coverage ratio that is greater than 0.6 for both financial statements. However, four of the five balance-sheet ratios are greater than 0.5 (except CE) and four of the five income-statement ratios are about 0.5 (except SIPP). The key differences across surveys occur in both types of coverage ratios for the balance sheets. The SCF has nearly complete value coverage of the balance sheet (above 0.9 by value) and the HRS has a value ratio about 0.8 (by value). Most surveys have item-coverage ratios of about half of the balance-sheet line items except the SCF, which covers the vast majority of line items. Variation across surveys is less in the item-coverage ratios for income statements.

C. Quantifying Integration by Dynamics

We also wish to characterize the degree to which surveys are integrated with household financial statements in terms of dynamics. Our proposed criterion for measuring this kind of integration is a quantification of the extent to which the estimated stock-flow identity holds in the survey estimates of household financial statements. The statement of CF is well suited to quantifying this measure of integration because it provides the linkage between the income statement (flows of income and expenditures) and changes in the balance sheet (stocks of assets and liabilities), assuming all stocks and flows are measured exactly and comprehensively. As explained in Section III, however, the CF error that arises in practice quantifies how well the balance sheet and income statement are integrated over time. CF errors represent consequences of incomplete item coverage of financial statements, as well as various forms of mismeasurement of the items in the financial statements.

Table 4 reports estimates of the statements of CF for each survey. Starting with net income (from the income statement), the estimated change in CF is the sum of three types of CF: from production, from consumption and investment, and from financing. To construct these statements, we have to estimate the elements of the CF from financing using estimated changes in the relevant assets and liabilities from the prior-period balance sheet. This methodology produces a CF estimate that is a residual difference between net income and net CF, rather than a direct measure of the gross CF in and out of the balance sheet, because the latter are not available from the U.S. surveys. For comparison, we estimate the change in cash holdings directly from the current and prior-period balance sheets. (32)

The degree of dynamic integration is defined as the difference (error) between the estimated CF variables and the change in cash holdings estimated from the current and prior period balance sheets, expressed in dollar terms and as a percentage of the lagged stock of cash. We call this the "internal" CF error because it is calculated using only the survey's estimates of stocks and flows. However, cash holdings from any particular survey may differ from the actual aggregate U.S. estimate of cash holdings (from the Flow of Funds), so these errors may not accurately represent the true degree of integration. Therefore, we also include the change in household cash holdings from the Flow of Funds (same for each survey) and construct errors in the survey CF estimates relative to the actual Flow of Funds cash to give a better measure of dynamic integration. We call this the "external" CF error.

As measured by their ability to track stock-flow identities in the statements of CF, the U.S. surveys exhibit relatively weak dynamic integration, and the degree of integration varies widely across surveys. The absolute value of the internal CF error ranges from $6,290 (CE) to $47,404 (SCF). Note that these errors are just one estimate in a time series of errors that could be estimated, and other errors might be smaller in absolute value during other periods. However, the sheer magnitude of these internal errors suggests significant gaps in tracking household financial conditions over time, even within the self-contained estimates of a particular survey. (33) The CF errors are reported in percentage terms relative to the two benchmarks: (1) the lagged cash stock from the survey's balance sheet (internal error); and (2) the lagged cash stock from the Flow of Funds aggregate benchmark data (external error). The internal errors are relatively large, ranging from about 13% to 37% of lagged cash (CE and SCF, respectively). The survey estimates of CF are generally less than the external benchmark: all but one of the external CF errors are even larger in absolute value, ranging from about 8% to 61% of lagged cash.

V. THE TTMS AND DCPC

Moving beyond the U.S. household surveys, we now focus on two other surveys that offer improved integration with financial statements and reflect better measurement of certain aspects of household economic conditions. The TTMS and DCPC are quite different in most regards. The TTMS is a comprehensive survey of household economic conditions, including home businesses; it is administered to Thai households, which are relatively low-income, less-developed, and located in rural geographic regions. In contrast, the DCPC is a relatively narrow consumer survey that is administered to U.S. consumers and is focused on payment choices. Nevertheless, the TTMS and DCPC both embody certain elements of improved integration with financial statements. The TTMS is heavily focused on the most basic and liquid M1 portions of "cash" (or current assets). The DCPC includes currency and is unique in this respect among the U.S. surveys that we analyze here. The DCPC also features other means of payment, for example, payments that use deposit accounts, although it does not track the level of these deposits.

This section compares and contrasts the TTMS and DCPC surveys. First, we present estimated balance sheets and income statements for each survey and discuss their degrees of integration by item coverage. Next, for each survey, we describe the methodology for measuring CF. Finally, we assess its degree of integration by dynamics, emphasizing its relatively high integration compared with the U.S. surveys. For this section, we combine survey responses from the DCPC with responses from the SCPC because both surveys are needed to estimate the financial statements as thoroughly as possible. For simplicity, we refer to the combined DCPC and SCPC estimates as "CPC."

A. Balance Sheets and Income Statements

Balance sheets and income statements constructed from the TTMS and CPC surveys appear in Tables 5 and 6, respectively. These statements are designed and organized similarly to the analogous statements from the U.S. surveys, with a few exceptions. In these tables, the TTMS and CPC data represent exactly the same time period (October 2012), and the TTMS estimates have been converted to U.S. dollars using the Thai baht exchange rate for October 2012. Unlike the U.S. survey entries, the entries are not annualized because both the TTMS and the DCPC are designed to be monthly surveys.

In general, the TTMS and CPC financial statements are not really comparable due to the relative magnitudes of their respective economies. The average asset value (Table 5) for TTMS households includes several types of business assets, and is $89,082, and the average asset value for CPC households is $301,425; this measure does not include any business assets. This difference is magnified by the fact that the CPC estimate is well below the highest estimate in the U.S. surveys (Table 2A) because it does not include any current assets beyond currency and approximates tangible assets only roughly. The average liability value is only $5,317 for TTMS households but, at $120,689, is more than 20 times larger for the CPC because there are relatively few borrowing options for Thai households. The disparity between the Thai and U.S. economies is even more evident from the income statements, shown in Table 6, where the average CPC household income is roughly three and one-half times larger than the average TTMS household income ($5,921 vs. $1,643), and nearly five times larger net of expenditures ($4,081 vs. $830).

One similarity between the TTMS and CPC financial statements is the predominance of currency among current asset holdings. The average TTMS household is estimated to have $30,874 in currency and less than $5,000 in other current assets (mostly bank accounts). The average CPC household has $836 in currency, which is the only type of current asset data collected. Although currency holdings are much lower in U.S. households than in Thai households, the other U.S. surveys (except the SIPP) estimate bank account holdings of about the same magnitude as Thai cash holdings, which are roughly $30,000, as shown in Table 2A. The improved 2015-2016 CPC also contains bank account balances (see below). The accuracy of the data on currency holding in Thai households could be improved, and we come back to this later.

In addition to differences in their respective economies, the TTMS and CPC survey instruments are sufficiently different to inhibit meaningful comparisons. The TTMS aims to collect data on all aspects of Thai household economic behavior, an aim that produces extensive estimates of the line items in the financial statements despite lower economic development. In contrast, the CPC strives to measure payments activity comprehensively and does not aim to cover financial-statement line items widely. For these reasons, comparisons of line-item coverage ratios between these surveys are not meaningful, nor are comparisons with the U.S. surveys.

B. Measuring Cash (Currency) Flows

TTMS Survey Instruments. ST apply this household financial accounting framework to households in the TTMS and create the accounts from a baseline 1998 comprehensive survey and then month-by-month interviews, currently up to month 205 and counting: that is, they have 17 years of monthly data. There was an initial enumeration of all structures and all households living in a village (or in an urban neighborhood), a census including who is eating and sleeping in what structure, and a description of family relationships across the individuals in these structures. The initial survey was an extensive baseline, measuring not only initial assets and liabilities, but also contracts and relationships, for example, borrowing and labor arrangements. There are month-by-month follow-up interviews with separate modules for assets and liabilities and for revenues and expenses of various production activities. Every transaction is measured in principle, subject to recall, for example, recall of purchases, sales, gifts, and labor supply. A key to implementing this large survey is the creation of rosters, lists of individuals in the household, debts not yet repaid, plots of land under cultivation, and so on, so that enumerators know which questions to ask.

The TTMS asks households for every transaction, such as a purchase, whether it was done in cash (currency), in kind, or as a gift. Again, the period of recall in the survey is the previous month (more exactly, the time since the last interview, which is roughly 30 days). Interviewers do not observe or ask about initial levels of cash holding, but they do try to measure these flows by assuming that the initial cash holding at the beginning of the survey was high enough so that households never run out of cash; that is, cash levels can go to zero but are never negative. Cash holding does hit the zero bound when households purchase a durable or investment good with cash, which is reassuring.

In contrast with this finding, ST infer that on average households hold relatively large cash positions. This leads to two related concerns. First, consumption expenditures in cash may be underestimated. In this case, double-entry bookkeeping hits with a vengeance in the sense that there could be two errors: an underestimate of cash consumption and an overestimate of cash on the balance sheet. Second, households may choose to underreport deposits into and withdrawals from savings accounts, although they typically do confirm many transactions, large and small. In this case, two items on the balance sheet, although offsetting, may be mismeasured.

In addition, because currency is not only a means of payment but also a store of value, it constitutes a relatively large portion of a household's wealth, on average. Therefore, households are understandably reluctant to report to enumerators how much currency they are holding. A second problem is the frequency of interviews, hence 30-day periods of recall. One potential remedy would have been to have households keep diaries of daily transactions for the entire month, or to use intensive diaries for shorter time intervals per respondent (as the DCPC does) to obtain a measure of aggregate activity. Initial attempts to implement a diary in real time at the request of the households themselves show great promise in dealing with this second problem. We may not know the initial balance (still hidden), but the changes in balances due to better-measured monthly transactions are more accurate. This is a step toward the degree of accuracy of the CPC surveys described below.

At the time of the conception and initiation of the TTMS in 1997, the use of payment devices other than cash was rare in these rural areas. Over time, there has been an increase in card dissemination and small levels of use. The TTMS was modified to incorporate cards into the survey, but measurement has been difficult due to many complex issues, including question design, accounting methods, tracking card payments, reconciling end-of-month statements, separating interest from principal, rolling over debt, and so on. The remainder of the paper describes the Boston Fed's DCPC, an approach that might have improved the TTMS, and then shows how the integrated financial accounts can be extended with the DCPC data to include multiple means of payment.

CPC Survey Instruments. The 2012 SCPC and 2012 DCPC are related but independent instruments that were implemented around October 2012 with a common sample of respondents from the RAND Corporation's American Life Panel (ALP). The SCPC is an approximately 30minute online questionnaire that collects data on consumer adoption and use of bank accounts and payment instruments. The DCPC is a 3-day mixed-mode survey with daily recording of payments in a paper memory aid (or other form) plus three daily online questionnaires to input memory-aid data plus answer additional questions based on recall within the day. In 2012, most respondents took the SCPC before their randomly assigned 3-day period during October, but some respondents completed the SCPC after the DCPC. The order did not affect survey responses because the instruments are independent.

Cash holdings (stock) data are collected by the SCPC and DCPC, which are related but distinctly different types of survey instruments, as described in Section II. The SCPC obtains estimates of cash held by respondents on their person ("pocket, purse, or wallet") or on their property (home, car, or elsewhere). (34) The 2012 DCPC obtained estimates of currency (no coins) held by respondents on their person on each of the four nights of the diary, asking the respondent to report amounts by denomination of the bills ($1, $2, $5, $ 10, $20, $50, and $ 100) and in total (summed for them in the online questionnaire). (35) In October 2012, U.S. holdings of currency on person were on average $56 per person with a median value of $22.

CF--deposits and withdrawals (payments)--are collected by the SCPC and DCPC as well. With regard to cash withdrawals made for expenditures (payments), the SCPC obtains estimates of the number of cash payments "in a typical period [week, month, year]," whereas the DCPC more precisely obtains estimates of the number and value of each cash payment (expenditure) made during a 3-day period. Both the SCPC and the DCPC collect data on the number and value of cash withdrawals from bank accounts and other sources. However, because cash withdrawals are relatively rare for most consumers, the DCPC does not obtain estimates that are as comprehensive for individual consumers as does the SCPC, which asks for "typical" currency withdrawals during a longer time period than 3 days. Only the DCPC tracks currency deposits to bank accounts and other sources plus other unusual currency activity (conversion of currency to/from other assets, exchanging coins for bills, and such).

Two additional differences between the SCPC and DCPC have important implications for their cash data. First, while both surveys ask respondents to record their cash holdings at the time of the survey, the SCPC allows respondents to estimate their holdings, while the DCPC requires respondents to count their cash on person (bills only, no coins) by reporting the number of bills of each denomination, and the online DCPC questionnaire assists respondents in summing the value of their cash holdings. As a result, the SCPC cash holdings data exhibit more rounding (to the nearest $5, $10, or $20) and approximation than the DCPC data. Second, the SCPC collects data on cash payments based on respondents' recall of their typical behavior, while the DCPC collects data that respondents record in essentially real time at the point of payment. Recall-based estimates of payments are likely to be inferior to recorded estimates due to potential errors from memory loss and time aggregation. For more information about the DCPC and its advantages in measuring consumer expenditures, see Schuh (forthcoming).

Measurement by Recall Versus Recording. By way of summarizing the material in this paper so far, we describe the main advantage of TTMS over the U.S. surveys and the innovation in the DCPC relative to the TTMS. The main advantage of TTMS is that it aims to achieve complete integration with household financial statements by line-item coverage and by stock-flow dynamics. To see this point, consider the following illustrative system of equations that reflects the subset of TTMS financial-statement estimates for the CF dynamics of Ml liquid assets:

[mathematical expression not reproducible]

where the two assets, k = {1,2}, are currency (1) and demand deposits (2) and [eta] denotes a composite measurement error. An overhead circumflex ("hat") denotes a variable that is estimated directly by the survey (TTMS). The exception is that the TTMS does not directly collect cash holdings every period, unlike the DCPC. Instead, the TTMS makes an estimate of the initial stocks, [mathematical expression not reproducible] and then uses these stock-flow identities to impute the estimates of cash stocks in subsequent periods, denoted by an overhead tilde (~). In the imputation procedure, the TTMS enforces the constraints imposed by the principles of integration, such as [[??].sub.kt] [greater than or equal to] 0, and makes judgmental adjustments where necessary.

Conceptually, the TTMS is fully integrated. It achieves complete integration by line-item coverage because it estimates all items of the balance sheet ([A.sub.1t], [A.sub.2t]) and CF statement ([D.sub.1t], [D.sub.2t], [W.sub.1t], [W.sub.2t]). As a result, the TTMS would also achieve complete integration by dynamics, provided it covered 100% of the dollar values of the items; in this case, the stock-flow dynamics would hold without error. However, it is essentially impossible for a survey to reach complete value coverage, due to sampling errors, among other challenges. For this reason, the TTMS imputes the periodic stock of currency using a judgmental estimate of the starting value of currency holdings for each household and adjusts it periodically if the stock-flow law of motion produces an invalid level estimate. Of course, the TTMS cannot claim to achieve full integration by dynamics or by item coverage in terms of dollar value, as TTMS estimates likely have measurement errors, as all surveys do. Nevertheless, the TTMS is generally much more integrated than the U.S. surveys analyzed earlier, which have much less than full integration by coverage (item or value) and relatively large errors in CF dynamics. The links between the income statement and the balance sheet were not incorporated into these U.S. surveys.

In particular, one type of measurement error likely occurring in the TTMS CF estimates arises from recall-based low-frequency (monthly) estimates of CF. As noted, recall errors may occur from memory loss due to time aggregation over the days of the month or over the number of cash deposits and withdrawals (payments). To see this, note that monthly currency withdrawals,

[W.sub.1t] = [[D.sub.t].summation over (d=1)] [[K.sub.t].summation over (k=1)][W.sub.1kdt],

are the sum over all opportunities and days, where 28 [less than or equal to] [D.sub.t] [less than or equal to] 31 and [K.sub.t] [greater than or equal to] 0. Like most U.S. surveys, the TTMS obtains an aggregate recall-based estimate of monthly cash withdrawals, [[??].sub.1t], from deposits to currency, without measuring each individual cash withdrawal, [W.sub.1kdt]. The same measurement issue holds for currency deposits, which are less frequent and thus may be measured with less error.

By comparison, daily payment diaries like the DCPC represent an innovation in the measurement of stock-flow dynamics by recording high-frequency (daily) CF. For example, the DCPC obtains an estimate of each individual cash withdrawal, [[??].sub.1kdt], by type, so the DCPC estimate of aggregate monthly cash withdrawals is the sum of individual withdrawals estimates,

[[bar.W].sub.1t] = [[D.sub.t].summation over (d=1)][[K.sub.t].summation over (k=1)] [??],

denoted by an overhead line. Therefore, if high-frequency (daily) recorded estimates of CF are more accurate than low-frequency (monthly) recall-based estimates, then we expect that

[absolute value of [bar.[W.sub.1t]] - [W.sup.*.sub.1t]] < [absolute value of [??] - [W.sup.*.sub.1t]],

at least on average, if not period-by-period as well. Consequently, the DCPC estimates of the stock-flow law of motion for currency,

[DELTA] [A.sub.1t] = [bar.[D.sub.1t]] - [bar.[W.sub.1t]] + [[mu].sub.1t],

are likely to be a better measure than those from the TTMS for the reasons enumerated above: (1) DCPC estimates of monthly currency flows are sums of individual opportunity-day flows; and (2) DCPC estimates of currency holdings are obtained each period, not derived from an initial condition (estimate) using the estimated flows. In this sense, the DCPC estimates improve the integration of surveys with financial statements and offer the opportunity for enhanced analysis of household behavior, as demonstrated below.

C. Statements of CF

The statements of CF constructed from the TTMS and CPC surveys appear in Table 7. In most respects, these CF statements are designed analogously to the statements of CF from the U.S. surveys (Table 4), and the elements are defined similarly to those in the balance sheets and income statements for TTMS and SCPC/DCPC (Tables 5 and 6). One exception is that the TTMS and DCPC represent CF and balance-sheet changes for one exact month (October 2012) rather than annual (or lower-frequency) flows. Also, bear in mind that the TTMS CF from financing equal the actual changes in the balance-sheet stocks. Therefore, the estimated change in currency from the CF statement equals the change from the balance sheet by definition; hence, the CF error is exactly zero because the stock-flow principle of motion is an identity, a significant step forward. Thus, the TTMS appears fully integrated by dynamics, but this integration is "artificially" high because it is derived rather than estimated directly.

CF in Thai and U.S. households differ in both magnitude and type. Net income is naturally much larger, $5,767 versus $729, in U.S. households. Adjustments to net income for accrual-based income in the statements of CF are modest for Thai households that have business income (a total increase of $ 130), and not measured for U.S. households ($0), so the difference in CF from production are still large, $5,767 versus $859. However, CF for consumption and investment by U.S. households are very large, estimated at $6,767, relative to net income but much smaller relative to income, estimated at $327, for Thai households. Similarly, U.S. CF from financing are larger, $259 versus $13, and more diverse, notably with respect to credit cards (which were not included in the 2012 TTMS). The estimated changes in currency from CF are roughly similar, $-741 versus $544, despite larger differences in net income and other flows. Finally, the CF error analysis is not relevant or comparable. The TTMS error is zero ($0) by definition because the balance-sheet changes are restricted to equal the CF. In contrast, the DCPC error is a legitimate derivation from estimates of all components of the stock-flow relationship. However, the error, $905, is relatively large, 135% of lagged currency, because the DCPC was not designed or implemented in a way that would ensure full dynamic integration. Instead, the DCPC calculations illustrate the potential advantage of a payment diary in tracking the gross flows of currency and the stock-flow dynamics in financial statements.

VI. AN INNOVATION TOWARD BETTER INTEGRATION

This section introduces an innovation to CF accounting that demonstrates a second advantage of the DCPC for moving another step toward complete ST integration of surveys and financial statements. The previous section explained how payment diaries like the DCPC produce better estimates of CF and stocks than monthly surveys do. In addition, payment diaries can produce estimates of CF that directly link individual asset and liability accounts to CF via the payment instrument, rather than just linking aggregate categories of assets and liabilities to aggregate categories of CF. The remainder of this section describes the linkage between the balance sheet and payment instruments and then presents a new analysis of CF by account, before concluding with a preview of further innovations in the 2015 DCPC.

A. Payment Instruments and Balance-Sheet Accounts

Table 8 depicts the linkage between payment instruments and their associated balance-sheet accounts: assets and liabilities. Payments are funded (settled) by one of two broad types of accounts: money (asset) and credit (liability). Money includes transactions balances, or M1 (currency plus checking accounts), plus certain non-transaction balances, which are part of M2. The latter are savings, but in some cases can support a limited number of payments directly from or to the account (account-to-account, or A2A, transfers). Payments funded by money are usually settled instantly (with cash) or with delays of at most a couple days. Alternatively, credit accounts fund payments that are settled much later; nonrevolving credit accounts (charge cards) require consumers to repay their debt during a certain period (typically a month), while revolving credit accounts (credit cards) offer consumers the option of rolling over some of the debt (up to a credit limit) to the future indefinitely in exchange for incurring interest charges. Monetary assets and unused credit limits are the liquidity that fund payments that are tracked by instrument in the DCPC. (36)

The linkage between payment instruments and balance-sheet accounts merits additional discussion before moving ahead. Table 8 reveals that in U.S. household balance sheets the linkage is not one-to-one, due to the proliferation of accounts and payment instruments in the U.S. monetary and payment system. This linkage complexity is most evident in the variety of instruments that can access various types of deposit accounts (including saving accounts in M2). In particular, debit cards, various types of checks, and electronic banking methods (OBBP and BANP) all can be used to authorize payment or transfer from different types of accounts. In addition, the linkages depicted in Table 8 reflect aggregation of individual accounts within a type of account that the overall pattern does not reveal. For example, the 2012 SCPC indicates that 38% of U.S. consumers have more than one demand deposit (checking) account (DDA), and 57% of consumers with multiple DDAs have multiple debit cards, typically one (per account holder) for each DDA. Consequently, the linkages between accounts and instruments can be disaggregated further to match specific accounts and instruments within the categories of Table 8. For example, a consumer (or household) may own two DDAs with a debit card for each; thus, it would be necessary to link DDA #1 to debit card #1, and similarly for the other account and card. The 2012 DCPC accurately measures the linkages between types of accounts and types of instruments (such as DDAs and debit cards), but it does not measure the linkages between specific individual accounts and specific individual instruments.

B. CF by Account

Given the linkage between accounts and instruments, the DCPC can also link balance-sheet accounts (or types of cash stocks) to household expenditures on consumer nondurable goods and services (or types of withdrawal flows). (37) Theoretically, a payment diary could link balance-sheet accounts for household capital goods to payments for investment in durable goods, but the 2012 DCPC did not track these concepts. In any case, the payment instrument plays the pivotal role because, for each payment, it directly links the balance sheet--that is, the asset or liability funding the payment--to consumer expenditures broadly defined (more broadly than narrow consumption) for each payment transaction.

Our major innovation of this paper is the "Statement of Account Flows," which is constructed using the DCPC and appears in Table 9. The rows in this new type of financial statement are generally formatted as in a statement of CF, but separately for each payment account. For example, the first column is the statement of currency flows, which records the inflows and outflows of currency for each type of transaction, starting with currency inflow from production activities (monthly basis) in row A and followed by currency outflow from consumption and investment activities in row B (separating consumption expenditure in row B.1 from capital expenditure in row B.2). Next, row C and its subsidiary rows report the net currency flows from financing activities and its components: deposits (inflows; the C.1 rows) of currency from each other account (DDA, nonfinancial deposit accounts [NFDA], foreign currency, long-term financial assets [LTFA], revolving debt, and other debt) and withdrawals (outflows; the C.2 rows) of currency to each of those accounts. The remaining rows compare the changes in currency balances from the statement of currency flows above (row D) with those estimated from the balance sheet (row E), plus an estimate of the error (in value and percentage of prior-period balance, rows F and G, respectively).

Similarly to the statement of currency flows in the first column, the remaining columns of the table represent information for the flows of DDA, NFDA, foreign currency, LTFA, revolving debt, and other debt, with the final column reporting the row sum. This provides the link from aggregate cash to each of the payments mechanisms. Importantly, note that the total net flows concept in row C appears in the last column ("All") as exactly zero by construction, since what goes into one payment account comes from another.

Total average account balances of U.S. consumers declined $1,004 in October 2012, according to the DCPC, as average consumption, at $6,771, exceeded total account flows from production activities, which were $5,767. This change in account balances tabulated from account flows resulted from much larger gross inflows and outflows, as withdrawals, at $8,524, exceeded deposits, which were $7,520. However, the decline in account balances estimated from the statement of account flows was considerably smaller in absolute value than the corresponding change estimated from balance-sheet stocks, which was $8,816. Therefore, the statement of account balances suggests that the DCPC is likely incomplete and may have considerable measurement errors, despite its conceptual promise for better integration by dynamics. One obvious area of incompleteness in the statement of account flows is that deposits of income to DDAs are not measured directly, but rather assumed to equal the difference between net income and currency deposits to income. (38)

The statement of account flows exhibits at least two interesting results with economic implications that may be useful for future research to link real (consumption) and nominal (financial) household choices. First, 99% of consumption, at $6,771, is funded by payments from DDAs (65.3%), from credit cards (18.4%), and from currency (15.3%). This result reflects heterogeneity in consumer payment choices, which may have implications for payment systems and for household budgeting and management of liquidity. Second, the gross-flow magnitudes are not small relative to income and consumption, which raises questions about the efficiency of the monetary system and relates to the classic literature on money demand: Why are U.S. households holding relatively large amounts of their liquid assets in payment accounts (just as Thai households hold so much in currency)? Also, it is still not entirely clear why consumers make such large transfers between currency and DDA, two assets that have the same monetary nature (M1) and are essentially equivalent for the settling of exchange. Evidence from the Survey of Consumer Payment Choice indicates that many U.S. consumers still rate the characteristics of currency (cost, speed, convenience, recordkeeping, and such) high relative to other payment instruments, and merchant acceptance of instruments is still not universal. Nevertheless, these large transfers between currency and DDA likely involve costs that may be reduced by the use of electronic money. All together, the account flows provide new data with advantages that potentially offer greater insight than existing data and research do into household financial decision making and the optimal design of the payments system more generally.

C. Improvements to the 2015 DCPC

While the 2012 DCPC introduced an innovation to the measurement of currency flows that has enhanced the degree of integration for one type of asset (currency), its coverage of financial statements has been relatively low, due to its limited mission and purpose. However, expanding the DCPC to measure the stocks of other assets from which consumers make payments not only increases coverage and integration but also provides important information for studying payment choices. For example, the analysis of the demand for currency and payment cards (debit and credit) by Briglevics and Schuh (2014) was limited by the lack of data on checking account balances. Also, the results by Schuh (2017) demonstrating the close correspondence between payments and personal income were produced without the benefit of direct measurement of the receipt of income by DCPC respondents.

Consequently, in 2015, the Boston Fed undertook to make major improvements to the SCPC and DCPC that substantially improved their integration with household integrated financial statements and the ST methodology. Improvements to the coverage of balance sheets included adding:

* Additional short-term liquid assets other than currency, including balances held in checking (DDA) and nonbank deposit accounts, such as prepaid cards, PayPal, and so on (SCPC and DCPC).

* Collection of outstanding debt balances from credit card bill payments (DCPC only).

Improvements to coverage of income and CF statements included adding:

* More intentional and detailed classification of expenditures based on official NIPA definitions of consumption, which increases the precision of the distinction between consumption and nonconsumption expenditures (DCPC only).

* Collection of the actual dollar values, types, and frequencies of personal income receipts, which will permit direct comparison of aggregate DCPC income with NIPA income (39) (DCPC only).

* Increased precision and information about the timing and nature of bill payments, which will improve the classification of expenditures and expand the capability to link payments to assets, and especially to liabilities (such as outstanding debt other than credit card debt).

Data from the 2015 and 2016 DCPC are in the process of being analyzed and prepared for publication in the near future.

D. Lessons for Survey Design

For all of the household financial surveys covered in this paper, and for any other similar survey, there is a relatively clear and straightforward path to developing complete integration with household financial statements. At least two main steps would need to be taken:

1. Obtain complete item coverage. All of the surveys are missing some line items from the balance sheet, income statement, or statement of CF. Adding survey questions to obtain estimates for each of these line items would provide complete item coverage. Of course, the coverage of a line item is not sufficient for full integration because errors may arise from sampling, question design, and other factors. Also, further disaggregation of the line items of the financial statements reported earlier may be required to achieve accurate aggregate estimates. Nevertheless, conditional on accurate estimation, comprehensive coverage of line items is a necessary step toward full integration. The surveys should also take into consideration innovations in financial instrument and payment methods, as they provide alternatives or replacements.

2. Ensure exact stock-flow identities. All surveys could improve the accuracy of their estimation of the dynamic identities inherent in the statement of CF. The use of high-frequency payment diaries appears to be one promising method for achieving this improvement. Provided the estimation of stocks (assets and liabilities) is relatively accurate, it is the estimation of aggregate flows (income and expenditures) over relatively long periods of time (minimum 1 month, but up to I year or more) that is the key survey methodology issue. Survey methods other than high-frequency payment diaries may yield improved estimates of aggregate flows, but it is not apparent which are the most successful. Further research is needed on this matter.

These two items are necessary for improving the integration of household financial surveys with household financial statements; they may also have interaction effects: for example, the omission of an asset from the balance sheet prevents improvements in the statement of CF. However, there may be other development issues to address as well, such as further improvements in the survey sampling frames.

VII. EXTENSIONS AND CONCLUSIONS

While the development issues necessary for integration are reasonably clear and straightforward, countervailing factors may inhibit comprehensive integration. One factor may be the lack of motivation, mandate, scope, or directive by the survey sponsors. Relatedly, the expansion of one survey may begin to overlap the coverage of another, which might be problematic for sponsors. For example, the SCF and CE each have relative strengths that, when combined, might move the collective dataset much closer to full integration of the accounts, but expansion of one or both of these surveys would create significant and costly duplication and would likely trigger a call for streamlining. Finally, an obvious inhibiting factor is the lack of sufficient budgetary resources to expand the survey and diary program, although budgetary resources are jointly determined with the previously mentioned factors.

The preceding discussion is equally relevant for the CPC survey and diary. Like all surveys, the 2012 SCPC and DCPC have advantages and disadvantages relative to the other surveys. However, one promising feature of the CPC survey and diary is that they have considerable room for quality improvements to the questionnaires that do not require additional budgetary resources, alternative sampling methods, or broader scope of operation and directive. The Boston Fed implemented the following improvements in the SCPC and DCPC during the fall of 2015, and the results will be forthcoming in future research.

* Separately identifying the payer (consumer) and payee rather than defining merchant categories that combine payee and type of expenditure, a separation that enables a far richer understanding of the purposes and reasons for the expenditure (including whether or not the expenditure was expected and the source of funding for unexpected expenses).

* Improvements to the statement of CF include additional information on how households finance their expenditures, and also provide additional real-time error-checking of online questionnaire responses, using stock-flow identities among assets, income, and expenditures.

These improvements highlight the fact that payment diaries link individual expenditure entries of the income statement with their associated assets and liabilities in the balance sheet and the detailed statement of CF in ways that have not been realized in other studies, including ST. However, the improvements are modest relative to the additional innovations that would be required to achieve complete integration, so much more research and data collection are needed.

The CE also is undergoing a redesign and improvement effort in response to recommendations from a National Academy of Sciences review panel, as described by National Research Council (2013). The report recommends considering three new prototype designs:

* Design A--Detailed expenditures through self-administration. This method would improve respondent reporting of expenditures and reduce respondent burden in data collection.

* Design B--A comprehensive picture of income and expenditures. This method would use technology, financial records, financial software, and budget balancing to improve estimates of the income statement.

* Design C--Dividing tasks among multiple integrated samples. This method would improve estimation of income-statement items through better use of sampling methodology.

While these improvements are valuable and promising, the NAS report does not appear to discuss or advocate the concept of integration beyond improvements to estimation of the income-statement line items.

A detailed discussion of research coming from the TTMS, SCPC, DCPC, and the other U.S. surveys is outside the scope of this paper. Many excellent contributions make use of each of the various surveys, and some use combinations of them. At the same time, analysts are limited in what they do without the integration of the accounts; indeed, a literature review would be useful to enumerate these strengths and limitations and to illustrate what might be done with improved data. Of course, this would take us well beyond the current endeavor.

Relatedly, although we have aggregated up to a common "representative" set of financial accounts, one would often like to disaggregate to some degree and go back to the underlying data organized by the accounts. Given the recent interest in the observed heterogeneous outcomes across U.S. communities in the lead-up and fallout from the financial crisis, it would be natural to disaggregate by geography (ZIP code, SMSA, commuting zone, county, and state). Unfortunately, many of the surveys were not designed to be representative at this level or lack sufficient observations to provide statistical significance. Indeed, one ends up taking one piece of data from one survey, another from another, and so on. But the available data are not organized systematically under the conceptual framework of integrated financial accounts. This, too, would seem to be a worthwhile endeavor that is beyond the scope of the current paper.

In the broader introduction to this paper and in the measurement efforts in the last few sections, we stressed the importance of payments data that could make it possible to distinguish among the payment instruments, align with more conventional measures of CF, and be used to calculate changes in balance-sheet items and income statements. Again, we have not had space in this paper to describe this connection in more detail. Suffice it to note that innovation in financial markets and monetary policy all point to issues related to the still-important use of currency and issues related to the potential of alternative media of exchange based on new asset accounts. Indeed, some papers in the literature already note that the impact of monetary policy as previously conducted was a function of the industrial organization of banks at a local level. In particular, the willingness and ability of households to substitute across cash and demand deposits was found to be crucial in gauging the impact of policy. Better data on payments is thus central to understanding the impact of monetary policy moving forward.

Although we have presented standard accounting practices, the measurement provided by the accounts should be consistent with the measurement suggested by theoretical models. For example, if there were complete markets for contingent claims, then future income flows would be conceptualized as discounted future income adding to contemporary wealth. Contingent assets lose value when the expected states of the world on which their value depends do not occur, but they gain in value if the contracted state is realized. Wealth or net worth would move only with aggregate shocks. With incomplete markets and contracts, it is easier to envision wealth as the buffer stock or pension fund used to deal with this uninsured uncertainty. In any event, there needs to be a review of the contracts and implicit understandings a household has entered into and scrutiny, in turn, of how to treat these in the accounts. This, as well, remains the subject of another paper.

ABBREVIATIONS

ALP: American Life Panel

BLS: Bureau of Labor Statistics

CAMS: Consumption and Activities Mail Survey

CCE: Cash and Cash Equivalents

CE: Consumer Expenditure Survey

CF: Cash Flows

CPI: Consumer Price Index

CPRC: Consumer Payments Research Center

DCPC: Diary of Consumer Payment Choice

DDA: Demand Deposit Account

HFCS: Household Finance and Consumption Survey

HRS: Health and Retirement Survey

ISIR: Institute for Social Research

LLC: Limited Liability Corporations

LSMS: Living Standard and Measurement Study

LTFA: Long-Term Financial Assets

NFDA: Nonfinancial Deposit Accounts

NIPA: National Income and Product Accounts

NORC: National Opinion Research Center

PSID: Panel Study on Income Dynamics

SCF: Survey of Consumer Finance

SCPC: Survey of Consumer Payment Choice

SIPP: Survey of Income and Program Participation

ST: Samphantharak and Townsend (2010)

TTMS: Townsend Thai Monthly Survey

REFERENCES

Agarwal, S., and W. Qian. "Consumption and Debt Response to Unanticipated Income Shocks: Evidence from a Natural Experiment in Singapore." American Economic Review, 104(12), 2014, 4205-30.

Ahmed, S., M. Imaduddin, A. Sumair, Quaratulein, and F. Naseem. "Quarterly Branchless Banking Newsletter." State Bank of Pakistan, Issue 15, 2015.

Briglevics, T., and S. Schuh. "This Is What's in Your Wallet ... and Here's How You Use It." Federal Reserve Bank of Boston Working Paper No. 14-5, 2014.

Campbell, J. "Household Finance." Journal of Finance, 61(4), 2006, 1553-604.

Carroll, C. D., T. F. Crossley, and J. Sabelhaus, eds. Improving the Measurement of Consumer Expenditures. Chicago, IL; London, UK: University of Chicago Press, 2015.

Chang, M., W. A. Darity Jr.. D. Hamilton. R. O. Jackson, M. Kim, and A. P. Munoz. The Color of Wealth in Boston. Boston: Federal Reserve Bank of Boston Joint Publication with The New School, 2015.

Crossley, T., and J. Winter. "Asking about Expenditures: What Have We Learned?" in Improving the Measurement of Consumer Expenditures, edited by C. D. Carrol, T. F. Crossley, and J. Sabelhaus. Chicago, IL; London. UK: University of Chicago Press, 2015.

Darity, W., D. Hamilton, A. E. Price, V. Sridharan, and R. Tippett. "Umbrellas Don't Make It Rain: Why Studying and Working Hard Isn't Enough for Black Americans." Insight Center for Community Economic Development Report, 2015.

Dillman, D. A., and C. C. House, eds. Measuring What We Spend: Toward a New Consumer Expenditure Survey. Washington, DC: National Academies Press, 2013.

Fulford, S., C. Greene, and W. Murdock III. "U.S. Consumer Holdings and Use of $1 Bills." Federal Reserve Bank of Boston Working Paper No. 15, 2015.

Greene, C., and S. Schuh. "U.S. Consumers' Holdings and Use of $100 Bills." Federal Reserve Bank of Boston Working Paper No. 14-3, 2014.

Greene, C., S. Schuh, and J. Stavins. "The 2014 Survey of Consumer Payment Choice: Summary of Results." Federal Reserve Bank of Boston Working Paper No. 163, 2016.

Imdieke, L. F., and R. E. Smith. Financial Accounting, 1st ed. New York: John Wiley & Sons, 1987.

Jack, W., T. Suri, and R. Townsend. "Monetary Theory and Electronic Money: Reflections on the Kenyan Experience." Economic Quarterly Federal Reserve Bank of Richmond, 96(1), 2010 (first quarter), 83-122.

Kochar. A. "Parental Benefits from Intergenerational Coresidence: Empirical Evidence from Rural Pakistan." Journal of Political Economy, 108(6), 2000, 1184-209.

Mian, A. R., and A. Sufi. "House Prices, Home Equity-Based Borrowing, and the US Household Leverage Crisis." American Economic Review, 101, 2011, 2132-56.

Mian, A. R., A. Sufi, and E. Verner. "Household Debt and Business Cycles Worldwide." Kreisman Working Paper Series in Housing Law and Policy. No. 38, 2015, revised 2017. Accessed June 2016. https://ssrn.com/abstract=2655804.

National Research Council. "Measuring What We Spend: Toward a New Consumer Expenditure Survey. Panel on Redesigning the BLS Consumer Expenditure Surveys," in Committee on National Statistics, Division of Behavioral and Social Sciences and Education, edited by D. A. Dillman and C. C. House. Washington, DC: The National Academies Press, 2013.

Samphantharak, K., and R. M. Townsend. Households as Corporate Firms: An Analysis of Household Finance Using Integrated Household Surveys and Corporate Financial Accounting. Cambridge, UK: Cambridge University Press, 2010.

Schuh, S. "Measuring Consumer Expenditures with Payment Diaries." Economic Inquiry, Forthcoming.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Appendix S1. Data Appendix

(1.) For example, Mian and Sufi (2011) study the aggregate impact of the home-equity-based borrowing channel and find that a large portion of total new defaults between 2006 and 2008 were from homeowners who had borrowed aggressively against the rising value of their houses. In a panel analysis of 30 countries, Mian, Sufi, and Verner (2017) find that an increase in the household debt-to-GDP ratio predicts lower GDP growth and high unemployment. Outside the United States, a study by Agarwal and Qian (2014) shows a negative consumption response by Singaporean households to a decrease in access to home equity, with the result concentrated in credit card spending and stronger among individuals with limited access to credit markets or with a high precautionary saving motive.

(2.) For more information on the HFCS, see https://www.ecb.europa.eu/pub/economic-research/research-networks/ html/researcher_hfcn.en.html.

(3.) Carroll, Crossley, and Sabelhaus (2015) contains numerous studies showing the various practical and theoretical tradeoffs inherent in attempting to use survey data to build economic aggregates, tradeoffs that can make comparing results from different surveys extremely challenging. For instance, Crossley and Winter (2015) note the difficulties survey designers can have even in defining the term "household," which can significantly affect the comparability of survey results. Similarly, surveys with a short reference period may underestimate infrequent purchases, while surveys with a long reference period may suffer from recall issues. Two surveys with different reference periods may have comparability issues.

(4.) See Ahmed et al. (2015) for more information on the rise of branchless banking in Pakistan.

(5.) Separately, Schuh (2017) reports that the DCPC produces estimates of U.S. consumer expenditures that greatly exceed those from the CE (and diary) and that approximately match National Income and Product Account estimates of comparably defined measures of consumption and disposable income.

(6.) For information about Federal Reserve efforts to stimulate innovations in the U.S. payment system, see https://fedpaymentsimprovement.org/.

(7.) For more information about the PSID, see https://psidonline.isr.umich.edu/.

(8.) For more information about the HRS, see http://hrsonline.isr.umich.edu/.

(9.) For more information about the CE, see http://www.bls .gov/cex/ and http://www.bls.gov/cex/esxovr.htm. The CE dates back to the 1800s but was not implemented annually until 1980; for details, see https://www.bls.gov/cex/ ceturnsthirty.htm.

(10.) For more information about the SCF, see http://www .federalreserve.gov/econresdata/scf/scfindex.htm.

(11.) For more information about the SIPP, see http://www .census.gov/sipp/.

(12.) The HRS includes consumers ages 50 years and older and thus includes households with relatively high income and assets, making it more representative of all U.S. consumers than other surveys that focus on subsets of the population, such as low-income consumers. Two nonrepresentative surveys merit analogous analysis but are not included here because they focus on selected low- and moderate-income (LMI) U.S. consumers. One is the U.S. Financial Diaries (USFD), produced jointly by the Center for Financial Services Innovation (CFSI) and the NYU Wagner Financial Access Initiative. For more information, see http://www .usfinancialdiaries.org/. Another is the National Asset Scorecard for Communities of Color (NASCC), which is very similar to the PSID. For more information, see https://socialequity .duke.edu/research/wealth. Darity et al. (2015), and Chang et al. (2015).

(13.) This conception of households as analogous to corporate firms raises some interesting issues. First, one may think of firms as registered corporate entities. But the financial accounts also apply to firms that are proprietorships, so formality or legality is not the issue, per se. More substantive complications remain. The first is how to treat membership in a household, not only with respect to changes due to births and deaths of family members but also with respect to changes due to marriages, divorces, and migration. For that matter, even within the family there may be individual ownership of assets and liabilities, traceable in principle when the distinction is clear to the family members, but often it is not. Or, in the other direction, seemingly separate families may in fact be closely related, not just by blood or marriage but also by financial transactions and behavior. This is the case for family and extended networks, as typically occurs in developing economies, but also in some advanced economies, such as Spain.

(14.) There are two further qualifications. First, there is an adjustment for net incoming unilateral transfers (e.g., gifts and remittances), which are not thought to be part of the return on investment projects per se but rather a financing device or even good will. These are not uncommon for households. Second, the balance sheet can change with asset appreciation or depreciation if these capital gains or losses are recognized in the income statement. Thus, it is easy to measure savings poorly if appreciation and depreciation change the balance sheet and income statements if one does not consider active flows of funds. Appreciation and depreciation can contribute substantially to increases and decreases in income, especially for those with substantial financial portfolios, as is the case for some older households.

(15.) Accrual-basis accounting, where revenues (income) are reported when they are earned and expenses (expenditures) are reported when revenues are reported, may be a more accurate representation of a company's net profits or financial condition (and a household's financial condition) than cash-basis accounting. Accrual-basis estimates would involve a substantial change. ST does this for the TTMS data, and the contrast of cash basis with accrual basis has been quite useful in research, as noted earlier. Note that the differences between cash basis and accrual basis become less relevant with annual data (in comparison to monthly or quarterly) since cash received and revenues recognized are likely reported in the same period (although some differences persist in the Thai data). Likewise, in such cases, cash outflows and expenses likely take place in the same period. These two accounting approaches are also less relevant for non-business households, whose incomes are less likely to involve inventories and trade credits. Another reason a small difference likely exists between cash and accrued income in the U.S. data is that a large portion of income earned by households in the United States is from wages, whose receipt mostly corresponds to the period when labor services are provided (the main caveat is the complication on how pensions are treated, as mentioned above).

(16.) Currency could also refer to foreign currency, such as Euros, or even private virtual currency, such as bitcoin, but we abstract from these because the holdings of these currencies by U.S. households are small and their liquidity is less than that of sovereign currency.

(17.) Recent innovations in the U.S. payment system include nonbank financial companies that take deposits and make payments, such as PayPal and general purpose reloadable (GPR) prepaid cards, such as Green Dot, NetSpend, and Blue Bird. In some cases, these nonbank and/or nonfinancial companies act as an agent between banks and households and deposit the money they receive into bank accounts. However, tracking the actual location of these assets is difficult and is attempted only in the CPC due to its focus on payments. For most households, bank deposits are the main type of cash, but nonbank deposits are becoming more common for some households, especially unbanked and lower-income households.

(18.) Some CF statements focus on "current assets," which is CCE plus other assets that can reasonably be expected to be converted into cash (or cash equivalents) within about a year. Some current assets are primarily attributable to business activity, which is not in the scope of U.S. financial surveys or covered well by them and is therefore excluded. These assets include accounts receivable, inventories, marketable securities, prepaid expenses, and other liquid assets. In theory, these items apply to household finance, but it would require significant changes in the scope and methodology of the U.S. surveys to include them.

(19.) ST also included deposits at financial institutions and rotating savings and credit association (ROSCA) positions in their balance sheets. However, these assets are not used much as a medium of exchange by TTMS households and they change little over time, so they were excluded from the definition of "cash." Nevertheless, the ST statements of CF include adjustments for changes in these other liquid assets.

(20.) The material in this section draws heavily from Imdieke and Smith (1987).

(21.) This interpretation of the error is likely to be valid for a point in time, as in our analysis later in the paper. However, the error could be small in absolute value at any point in time by chance, so a better measure over time might be the average absolute error over time.

(22.) Assets and liabilities are owned by individual consumers, denoted by subscript i, who are members of a household. denoted by subscript h. Agent identifiers are suppressed for simplicity because the following discussion assumes aggregation occurs across all agents eventually.

(23.) The day-specific flows are net of intraday deposits and withdrawals, so this accounting could occur even more frequently (hourly or even by the minute) to obtain further insight into CF.

(24.) This discussion and conceptualization apply even if a survey does not have disaggregated data. Some notion of cash is implicitly being used. That said, one can imagine how errors could arise, in particular, discrepancies between the income statement and balance sheet.

(25.) This conversion is necessary because of differences in the sampling units. For surveys that do not use households as the reporting unit, we sum across all reporting units to get the U.S. total and then divide by a common estimate of the number of households from the March Current Population Survey (CPS).

(26.) This classification naturally involves some discretion as to the grouping and especially the level of aggregation. The latter affects the quantitative measure of integration later, but can be made higher or lower for alternative analyses.

(27.) We again thank the staff members of each survey program who did so. This comparison is painstaking and difficult for one survey, much less several, and it is a challenge even for the survey managers. Thus, we view our results in this section as preliminary and welcome further development and improvement of the analysis. To this end, we are making underlying data and software programs available to the public, and we invite other researchers to refine and expand our analysis.

(28.) There are some tradeoffs between using book value and market value. For illiquid assets (of any type) that are rarely traded, market value is not readily available. Subjective assessments of value are prone to have measurement errors. In such cases, conservative accounting practices value the assets at historical cost. In contrast, mark-to-market requirements may be more appropriate when markets are thick and volatility is not excessive.

(29.) Respondents to the SCF report actual currency holdings only if they choose to do so in an optional response about other assets, and this category also includes "cash" that is not currency, like prepaid cards. The SCF estimate is very small relative to the amount reported by Greene, Schuh, and Stavins (2016) from the SCPC, which indicates average total cash holdings per consumer of $207 (excluding large holdings, which represent the top 2% but are not estimated precisely).

(30.) For more information about these business structures and their tax implications, see https://www.irs.gov/ businesses/small-businesses-self-employed/business-structures.

(31.) The number of sole proprietorships and partnerships was equal to about 24% of U.S. households in 2012, and about 6% of U.S. employment is self-employment as of 2016. The actual share of households with one of these businesses depends on the type of business and the composition of households. but we lack sufficient data to make exact calculations.

(32.) The duration of the preceding period varies according to the frequency of the surveys, from one quarter (CE) to 3 years (SCF).

(33.) In principle, it would be interesting to compare the coverage ratios with the CF errors to quantify the relationship between them. However, with only one point-in-time estimate of coverage and dynamic integration for a handful of surveys, such an analysis would be premature. With more data on CF errors over time, it might be feasible to conduct such an analysis.

(34.) Measuring cash in "pocket, purse, or wallet" is an approximate method of identifying actual "transactions balances" of cash. Although it does not ask the respondent for these balances directly, it is a relatively objective and easy method of collecting these data. An alternative approach is to ask for "transactions balances" directly, as in the Survey of Household Income and Wealth in Italy (http://www.eui.eu/Research/Library/ResearchGuides/Economics/Statistics/DataPortal/SHIW.aspx). The SCPC also estimates U.S. consumer holdings of cash balances "on their property" (house, car, etc.), and some of this cash may be intended (eventually) for use in transactions as well. However, it is unclear whether respondents have an appropriate understanding of transactions balances or provide accurate estimates of them.

(35.) See Fulford, Greene, and Murdock (2015) for an analysis of $1 bills and Greene and Schuh (2014) for an analysis of $100 bills.

(36.) Note that deposits into an asset account are similar to reductions in loan accounts, although one is an asset and the other a liability. Likewise, withdrawals from an asset account are similar to increases in loan accounts. But there is a substantive difference in that asset accounts require deposits before being used, whereas liability accounts can be unfunded initially and repaid later.

(37.) If designed properly, a payments diary also could link balance-sheet accounts to the expenditures of household businesses, but we omit these from the discussion because the DCPC instructed respondents to exclude household business payments.

(38.) Furthermore, the income of individual consumers (2012 DCPC respondents) is not estimated directly. We use the 2012 SCPC estimate of household income for the respondent (reported in categorical form rather than in exact dollar amounts) and other data in the SCPC. DCPC, and SCF to impute income for the DCPC respondents. This shortcoming was partially addressed in the 2015 DCPC (see Section VI.C below).

(39.) The 2012 DCPC only asked for the days on which income was received by the respondent, not the dollar amount of income of individual respondents. The 2012 and 2015 SCPC asked for total household income in dollar ranges.

KRISLERT SAMPHANTHARAK, SCOTT SCHUH and ROBERT M. TOWNSEND *

* Stacy Carlson, Mi Luo. Jason Premo, Giri Subramaniam, and David Zhang provided excellent research assistance. We thank Allison Cole, Claire Greene, John Sabelhaus, Robert Triest, the editor of Economic Inquiry, and an anonymous referee for helpful comments, and Suzanne Lorant for excellent editing. We also thank all the staff members who manage the surveys studied in this paper for their assistance and reviews of our data calculations. The Townsend Thai Monthly Survey has been supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) grant number ROI HD027638; the research initiative PEDL, a program funded jointly by CEPR and DFID, under grant MRG002_1255; the University of Thai Chamber of Commerce; the Thailand Research Fund; and the Bank of Thailand. The views expressed in this paper are those of the authors and do not necessarily reflect the views of the Federal Reserve Bank of Boston or the Federal Reserve System.

Samphantharak: Associate Professor, Department of Economics, University of California San Diego, La Jolla, CA 92093-0519. E-mail [email protected]

Schuh: Director, Research Department, Federal Reserve Bank of Boston, Boston, MA 01887. Phone (617) 973-3941, Fax (617) 619-7541, E-mail [email protected]

Townsend: Elizabeth & James Killian Professor, Department of Economics, Massachusetts Institute of Technology, Cambridge, MA 02142-1347. E-mail [email protected]

doi: 10.1111/ecin.12489

Online Early Publication October 12, 2017

Caption: FIGURE 1 Relation between Household Income Statement and Balance Sheet

Caption: FIGURE 2 Constructing Financial Statements from a Panel Household Survey

Caption: FIGURE 3 Financial Statement Line-Item Coverage Ratios for U.S. Surveys

TABLE 1
Overview of U.S. Surveys and Diaries and TTMS

                     PSID               CE-S/D

Sponsor              University of      BLS
                     Michigan

Vendor               University of      Census Bureau
                     Michigan

Frequency            Biennial           Monthly

Period               1968-present       1980-present

Statistical          2011, 2013         2011, 2012
calculations

Questionnaires

Observation unit     U.S. family unit   U.S. consumer units

Mode(s)              Interview          Interview, diary

Data collection      Recall             Recording, recall

Measurement          Past year          Daily expenditures
period                                  (diary), or past year
                                        (survey)

Sampling

Target population    Total U.S.         Total U.S.
                     noninstitutional   noninstitutional

Sampling frame       Survey research    U.S. Census Bureau
                     center national    master address file
                     sampling frame

Sample size          ~10,000            ~7,000

Longitudinal panel   4 consecutive      14 days
                     quarters

                     SCF                  SIPP

Sponsor              Federal Reserve      Census Bureau
                     Board

Vendor               NORC/University of   Census Bureau
                     Chicago

Frequency            Triennial            Quarterly

Period               1983:Ql-present      1983:Q4-present

Statistical          2009, 2012           2010, 2011
calculations

Questionnaires

Observation unit     U.S. primary         U.S households
                     economic units

Mode(s)              Interview            Interview

Data collection      Recall               Recall

Measurement          "Average" week for   Past month, past
period               expenditures, past   4 months, or past
                     year for income      year

Sampling

Target population    Total U.S.           Total U.S.
                     noninstitutional

Sampling frame       NORC national        U.S. Census Bureau
                     sampling frame and   master address file
                     1RS data

Sample size          ~6,000               14,000-52,000

Longitudinal panel   None                 2.5-4 years

                     HRS/CAMS               S/D-CPC

Sponsor              University of          Boston Fed
                     Michigan

Vendor               University of          RAND/University of
                     Michigan               Southern California

Frequency            Biennial               Yearly/Irregular

Period               2008-present           2012, 2015

Statistical          2010, 2012              2011, 2012
calculations

Questionnaires

Observation unit     U.S. households        U.S. consumers and
                                            households

Mode(s)              Interview, mail        Interview, diary

Data collection      Recall                 Recording (1 day),
                                            recall (1 year)

Measurement          Past year              Daily payments
period                                      (DCPC), or
                                            "typical" week,
                                            month, year
                                            (SCPC)

Sampling

Target population    U.S. ages 50+          Age 18+
                     noninstitutional       noninstitutional

Sampling frame       Panel of adults born   RAND ALP, USC
                     1931-1941              UAS, GfK
                                            knowledge
                                            networks

Sample size          9,000-15,000           ~2,000

Longitudinal panel   Fixed                  3-day waves tied to
                                            SCPC annual panel

                     TTMS

Sponsor              Townsend
                     Thai
                     Monthly
                     sponsors

Vendor               Thai Family
                     Research
                     Project

Frequency            Monthly

Period               1998-present

Statistical          2012
calculations

Questionnaires

Observation unit     Thai
                     households

Mode(s)              Interview

Data collection      Recall

Measurement          Past month
period

Sampling

Target population    Rural and
                     semi-urban
                     households

Sampling frame       Initial village
                     census

Sample size          ~800

Longitudinal panel   1998-present

CE-S: http://www.bls.gov/CE/capi/2015/cecapihome.htm;
CE-D: http://www.bls.gov/CE/ced/2013/cedhome.htm; TTMS:
http://townsend-thai.mit.edu/about/; SIPP:
http://www.census.gov/programs-surveys/sipp/about.html;
PSID: https://psidonline.isr.umich.edu/; SCPC:
http://www.bostonfed.org/economic/cprc/scpc/;
DCPC: https://www.bostonfed.org/economic/cprc/
data-resources.htm; SCF: https://www.federalreserve.gov/-
econresdata/scf/scfindex.htm; HRS/CAMS:
https://hrs.isr.umich.edu/about

TABLE 2
U.S. Surveys: Balance Sheets: (A) Assets, Various
Dates and (B) Liabilities, Various Dates

                                     PSID       CES       SCF

(A) Assets                          422,616   226,314   632,246
  Median                            151,000             170,600
  Financial assets                  163,376   65,537    262,168
  (% of assets)                      (39)      (29)      (41)
    CURRENT ASSETS                  95,883    65,115    140,176
      Cash                          29,850    30,849    30,354
        Currency                                          12
          Government-backed                               12
          currency
          Private virtual
          currency
        Bank accounts               29,850    30,849    30,342
          Checking accounts                   17,239    12,660
          Savings accounts                    13,610    17,682
        Other deposit accounts                             0
      Other current assets          66,033    34,266    109,822
        Certificates of deposit                          4,994
        Bonds                                   408      8,227
        Mutual funds/hedge funds                        40,964
        Publicly traded equity      56,335    33,858    48,874
        Life insurance               9,698               6,763
    LONG-TERM INVESTMENTS           67,493      422     121,992
      Retirement accounts           67,493              97,007
      Annuities                                          5,490
      Trusts/managed                                    13,773
      investment accounts
      Loans to people outside                   422      5,722
      the household
      Other important assets
  Tangible (physical) assets        259,240   160,777   362,445
  (% of assets)                      (61)      (71)      (57)
    Business                        51,404              108,760
    Housing assets                  188,992   160,777   234,187
      Primary residence             149,211   149,760   170,159
      Other real estate             39,781    11,017    64,028
    Vehicles                        18,844              19,498
  Unknown assets                                         7,633
  (%) of assets)                                          (1)
(B) Liabilities                     82,288    73,668    112,306
  Median                            18,800              23,000
  Revolving debt                     2,671     4,512     2,185
  (% of liabilities)                  (3)       (6)       (2)
    Credit cards/charge cards        2,671     4,447     2,096
    Revolving store accounts                    65        89
  Nonrevolving debt                 79,617    69,156    110,121
  (% of liabilities)                 (97)      (94)      (98)
    Housing                         67,506    58,143    87,223
      Mortgages for primary         54,856    52,559    63,889
      residence
      Mortgages for investment      12,650     3,086    19,598
      real estate or second
      home
      HELOC/HEL                                2,498     3,556
      Loans for improvement                               180
    Loans on vehicles                4,310     3,926     4,508
    Education loans                  6,507               5,788
    Business loans                                      10,317
    Investment loans                                      289
    (e.g., margin loans)
    Unsecured personal loans
    Loans against pension plan                            288
    Payday loans/pawn shops
    Other loans                      1,294     7,087     1,708
Net worth (equity)                  340,328   152,646   519,940
  Cumulative net gifts
  received
  Cumulative savings

                                      HRS      SIPP

(A) Assets                          556,295   351,702
  Median                            240,000   67,113
  Financial assets                  205,461   160,651
  (% of assets)                      (37)      (46)
    CURRENT ASSETS                  125,898   102,642
      Cash                          34,733    12,434
        Currency
          Government-backed
          currency
          Private virtual
          currency
        Bank accounts               34,733      536
          Checking accounts                     536
          Savings accounts
        Other deposit accounts                11,898
      Other current assets          91,165    90.208
        Certificates of deposit      9,354
        Bonds                       14,860     3,376
        Mutual funds/hedge funds              18,830
        Publicly traded equity      66,951
        Life insurance                        68,002
    LONG-TERM INVESTMENTS           79,563    58,009
      Retirement accounts           79,563    54,759
      Annuities
      Trusts/managed
      investment accounts
      Loans to people outside                   361
      the household
      Other important assets                   2,889
  Tangible (physical) assets        336,951   191,051
  (% of assets)                      (61)      (54)
    Business                        55,006    25,921
    Housing assets                  264,500   154,795
      Primary residence             190,818   147,855
      Other real estate             73,682     6,940
    Vehicles                        17,445    10,335
  Unknown assets                    13,883
  (%) of assets)                      (2)
(B) Liabilities                     64,614    61,979
  Median                             5,600     3,750
  Revolving debt                               2,661
  (% of liabilities)                            (4)
    Credit cards/charge cards
    Revolving store accounts
  Nonrevolving debt                 64,614    59,318
  (% of liabilities)                 (100)     (96)
    Housing                         58,584
      Mortgages for primary         48,984
      residence
      Mortgages for investment       4,440
      real estate or second
      home
      HELOC/HEL
      Loans for improvement          5,160
    Loans on vehicles                          3,707
    Education loans
    Business loans                             5,338
    Investment loans                            102
    (e.g., margin loans)
    Unsecured personal loans
    Loans against pension plan
    Payday loans/pawn shops
    Other loans                      6,030    50,171
Net worth (equity)                  491,681   289,723
  Cumulative net gifts
  received
  Cumulative savings

Notes: Table entries are average dollar values for
the survey's unit of observation, approximately a
household. Assets and liabilities are stocks dated
as of the time of the survey, generally the end of
the year. Sampling weights provided by each survey
were used in calculating the average values in accordance
with the survey's data documentation. A more detailed data
appendix (Appendix SI, Supporting Information) and the Stata
programs used to construct the tables are available at
http://dx.doi.org/10.7910/DVN/F7JB 1K. HELOC/HEL,
home equity line of credit / home equity loan.

Sources: PSID 2013, CE 2012, SCF 2013, HRS 2012, and
SIPP 2011. See Section II for more details.

TABLE 3
U.S. Surveys: Income Statements, Various Dates

                                               PSID     CES

Income                                        67,187   65,316
  Median                                      44,500   46,774
  Labor income                                53,623   51,543
  (% of total income)                          (80)     (79)
    Wages and salaries                        53,473   51,543
    Professional practice or trade             113
    Other labor earnings                        37
  Production income                           3,748    3,075
  (% of total income)                          (6)      (5)
    Business income (self-employment)         2,472    2,926
    Rent                                      1,276     149
  Other income                                9,816    10,698
  (% of total income)                          (15)     (16)
    Interest, dividends, etc.                 2,206    1,204
    Government transfer receipts              1,302    5,812
    Other transfer receipts, from business     131
    Other transfer receipts, from persons               380
    All other income                          6,177    3,302
Expenditures                                  1,837    4,345
  Production costs
  (% of total expenditures)
    Depreciation
    Capital losses
    Business expenses
    Cost of labor provision
    Cost of other production activities
  Taxes                                       1,837    4,345
  (% of total expenditures)                   (100)    (100)
    Employment taxes                                   2,508
    Other taxes                               1,837    1,837
Net income                                    65,350   60,971

                                               SCF      HRS      SIPP

Income                                        83,863   79,779   61,431
  Median                                      45,000   46,300   45,396
  Labor income                                53,192   42,377   48,767
  (% of total income)                          (63)     (53)     (79)
    Wages and salaries                        53,192
    Professional practice or trade
    Other labor earnings
  Production income                           11,347            1,144
  (% of total income)                          (14)              (2)
    Business income (self-employment)         11,347
    Rent                                                        1,144
  Other income                                19,324   37,402   18,176
  (% of total income)                          (23)     (47)     (30)
    Interest, dividends, etc.                 6,682    18,093
    Government transfer receipts              10,670   12,415   7,294
    Other transfer receipts, from business              423
    Other transfer receipts, from persons      372
    All other income                          1.600    6,471    10,882
Expenditures                                  2,007      0      22,487
  Production costs
  (% of total expenditures)
    Depreciation
    Capital losses
    Business expenses
    Cost of labor provision
    Cost of other production activities
  Taxes                                       2,007             2,798
  (% of total expenditures)                   (100)
    Employment taxes                                             585
    Other taxes                               2,007             2,213
Net income                                    81,856   79,779   38,944

Notes: Table entries are average dollar values for
the survey's unit of observation, approximately a
household. Income and expenses are reported for the
prior 12 months, or annualized where necessary. Sampling
weights provided by each survey were used in calculating
the average values in accordance with the survey's data
documentation. A more detailed data appendix (Appendix S1)
and the Stata programs used to construct the tables are
available at http://dx.doi.org/10.7910/DVN/F7JB1K.

Sources; PSID 2013, CE 2012, SCF 2013, HRS 2012, and
SIPP 2011. See Section II for more details.

TABLE 4
U.S. Surveys: Statements of CF

                                      PSID         CES
(Cash Defined as Current Assets)    2010-2012   2011-2012

Net income (+)                       65,350      60,971
Adjustments:
  Depreciation (+)                      0           0
  Change in account                     0           0
  receivables (-)
  Change in account payables (+)        0           0
  Change in inventory (-)               0           0
  Change in other (not cash)            0           0
  current assets (-)
  Consumption of household              0           0
  produced outputs (-)
CF from production                   65,350      60,971
  Consumption expenditure (-)        -43,766     -44,849
  Capital (durable goods)               0           0
  expenditure (-)
CF from consumption and              -43,766     -44,849
investment
  Transfers to/from long-term         -362          0
  investments
  Leading (-)                           0         -151
  Borrowing (+)                       4,230       8,089
  Net gifts received (+)                0           0
CF from financing                     3,868       7,938
  Change in cash holding             25,452      24,060
  (from statement of CF)
  Change in cash holding              3,091      17,770
  (from statement of
  balance sheet)
CF error                             22,362       6,290
Internal error (%)                     25          13
External error (%)                     30           8

                                       SCF         HRS
(Cash Defined as Current Assets)    2010-2013   2010-2012

Net income (+)                       81,856      79,779
Adjustments:
  Depreciation (+)                      0           0
  Change in account                     0           0
  receivables (-)
  Change in account payables (+)        0           0
  Change in inventory (-)               0           0
  Change in other (not cash)            0           0
  current assets (-)
  Consumption of household              0           0
  produced outputs (-)
CF from production                   81,856      79,779
  Consumption expenditure (-)        -28,850     -45,073
  Capital (durable goods)               0           0
  expenditure (-)
CF from consumption and              -28,850     -45,073
investment
  Transfers to/from long-term         1,231         0
  investments
  Leading (-)                         1,359        50
  Borrowing (+)                      -4,349      -3,757
  Net gifts received (+)                0           0
CF from financing                    -1,759      -3,707
  Change in cash holding             51,247      31,000
  (from statement of CF)
  Change in cash holding              3,843       1,678
  (from statement of
  balance sheet)
CF error                             47,404      29,322
Internal error (%)                     37          24
External error (%)                     61          39

                                      SIPP
(Cash Defined as Current Assets)    2010-2011

Net income (+)                       38,944
Adjustments:
  Depreciation (+)                      0
  Change in account                     0
  receivables (-)
  Change in account payables (+)        0
  Change in inventory (-)               0
  Change in other (not cash)            0
  current assets (-)
  Consumption of household              0
  produced outputs (-)
CF from production                   38,944
  Consumption expenditure (-)        -22,487
  Capital (durable goods)               0
  expenditure (-)
CF from consumption and              -22,487
investment
  Transfers to/from long-term           0
  investments
  Leading (-)                         4,452
  Borrowing (+)                      -8,988
  Net gifts received (+)                0
CF from financing                    -4,536
  Change in cash holding             11,921
  (from statement of CF)
  Change in cash holding             -18,622
  (from statement of
  balance sheet)
CF error                             30,543
Internal error (%)                     25
External error (%)                     42

Notes: Table entries are average dollar values for the
survey's unit of observation, approximately a household.
CF are at a yearly rate and are constructed with the most
recent prior data available. Sampling weights provided by
each survey were used in calculating the average values.
A more detailed data appendix (Appendix S1) and the Stata
programs used to construct the tables are available at
http://dx.doi.org/10.7910/DVN/F7JBlK.

Sources: PSID 2010-2013, CE 2011-2012, SCF 2010-2013,
HRS 2010-2012, and SIPP 2010-2011. See Section II for
more details.

TABLE 5
TTMS and SCPC/DCPC: Balance Sheets, October 2012

                                              DCPC/
                                     TTMS     SCPC

Assets                              89,082   301,425
  Median                                     146,053
  Financial assets                  35,553     836
  (% of assets)                      (40)      (0)
    CURRENT ASSETS                  35,321     836
      Cash                          35,332     836
        Currency                    30,874     836
          Government-backed         30,874     836
          currency
        Bank accounts               4,458
      Other current assets           -11
        Certificates of deposit
        Net ROSCA position           -11
        Accounts receivable           0
        Bonds
        Mutual funds/hedge funds
        Publicly traded equity
        Life insurance
    LONG-TERM INVESTMENTS            232
      Retirement accounts
      Annuities
      Trusts/managed investment
      accounts
      Other lending                  232
  Tangible (physical) assets        53,529   148,421
  (% of assets)                      (60)     (49)
    Business assets                  334
    Agricultural assets             1,243
    Housing/household assets        4,582    148.421
      Primary residence                      148,421
    Inventories                     8,394
    Livestock                        290
    Other nonfinancial assets       38,687
  Unknown assets                             152,168
  (% of assets)                               (50)

                                                   DCPC/
                                          TTMS     SCPC

Liabilities                              5,317    120,689
  Median                                          42,935
  Revolving debt                                   5,306
  (% of liabilities)                                (4)
    Credit cards/charge cards                      5,306
    Revolving store accounts
  Nonrevolving debt                      5,317    115,383
  (% of liabilities)                               (96)
    Housing                                       67,278
      Mortgages for primary residence             67,278
      Mortgages for investment real
      estate
      HELOC/HEL
      Loans for improvement
    Accounts payable                     1,480
    Loans on vehicles
    Education loans
    Business loans
    Investment loans
    (e.g., margin loans)
    Unsecured personal loans
    Loans against pension plan
    Payday loans/pawn shops
    Other loans                          3,837    48,105
Net worth (equity)                       83,765   180,736
  Cumulative net gifts received
  Cumulative savings                     56,779

Notes: Thai baht converted to U.S. dollars at a rate
of 30.68 baht per dollar. Values are stocks as of the
time of the survey, which for the CPC is between the
beginning of September and the end of October. TTMS entries
are at the household level. CPC entries are either at the
household level or converted to a household level by
multiplying consumer values by 2.045. A more detailed
appendix (Appendix SI) and the Stata programs used to
construct the tables are available at
http://dx.doi.org/10.7910/DVN/F7JB IK. HELOC/HEL, home
equity line of credit / home equity loan.

Sources: TTMS and SCPC.

TABLE 6
TTMS and SCPC/DCPC: Income Statements, October 2012

                               TTMS    SCPC/DCPC

Income                         1,643     5,921
  Median                                 4,413
  Censored income                        4,789
  Labor income                  252
  (% of total income)          (15)
  Production income            1,368
  (% of total income)          (83)
    Business                    326
    Agricultural activities    1,042
      Cultivation               536
      Livestock                 392
        Produce                 390
        Capital gains            2
      Fish and shrimp           114
  Other income                  23
  (% of total income)           (1)

                                  TTMS    SCPC/DCPC

Expenditures                       813      1,840
  Production costs                 813
  (% of total expenditures)       (100)
    Business                       251
    Agricultural activities        529
      Cultivation                  133
      Livestock                    292
      Capital losses                1
      Depreciation                 12
      Other expenses               280
      Fish and shrimp              104
    Labor provision                32
    Other production activities     1
  Taxes                                     1,840
  (% of total expenditures)                 (100)
Net income                         830      4,081

Notes: Thai baht converted to U.S. dollars at a rate
of 30.68 baht per dollar. Values are stocks as of the
time of the survey, which for the CPC is between the
beginning of September and the end of October. TTMS
entries are at the household level. CPC entries are
either at the household level or converted to a
household level by multiplying consumer values by
2.045. CPC household income is originally reported
in buckets; precise estimates are imputed with the
help of SCF data. A more detailed appendix (Appendix S1)
and the Stata programs used to construct the tables are
available at http://dx.doi.org/10.7910/DVN/F7JBlK.

Sources: TTMS, DCPC, and SCPC.

TABLE 7
TTMS and DCPC: Statements of CF, October 2012

(Cash Defined as Currency)                         TTMS     DCPC

Net income (annual basis) (+)                      8,750   69,207
Net income (monthly basis) (+)                      729    5,767
Adjustments:
  Depreciation (+)                                  94       0
  Change in account receivables (-)                 -37      0
  Change in account payables (+)                     0       0
  Change in inventory (-)                           80       0
  Consumption of household produced outputs (-)     -6       0
  Net capital gains (+)                             -1
CF from production                                  859    5,767
  Consumption expenditure (-)                      -245    -6,767
  Capital (durable goods) expenditure (-)           -77      0
CF from consumption and investment                 -327    -6,767
  Change in demand deposits (-)                     -67     -421
  Change in NFDA deposits (-)                       NA       59
  Change in foreign currency (-)                    NA       -2
  Change in credit card balance (-)                 NA     1,292
  Change in long-term assets (-)                    76      -669
  Change in other debts (-)                          4       NA
CF from financing                                   13      259
Change in currency balance                          544     -741
(from statement of CF)
Change in currency balance                          544     164
(from statement of balance sheet)
CF error                                             0      905
Internal error                                      NA      135%

Notes: Thai baht converted to U.S. dollars at a rate
of 30.68 baht per dollar. Values are stocks as of the
time of the survey, which for the CPC is between the
beginning of September and the end of October. TTMS
entries are at the household level. CPC entries are
either at the household level or converted to a
household level by multiplying consumer values
by 2.045. CPC household income is originally reported
in buckets; precise estimates are imputed with the
help of SCF data. A more detailed appendix (Appendix S1)
and the Stata programs used to construct the tables are
available at http://dx.doi.org/10.7910/DVN/F7JBlK.

Sources: TTMS, DCPC, and SCPC.

TABLE 8
Payment Instruments and their Balance Sheet Accounts

Balance Sheet Accounts             Payment Instruments

Assets (money)                     U.S. currency, foreign
  Currency                         currency, private currency
                                          (e.g., Bitcoin)
  Traveler's check                 Traveler's check
  Checking accounts owned by       Checks (personal or certified),
  consumers (demand and other      debit card, OBBP, BANP
  checkable deposits)
  Checking accounts owned or       Cashier's check, prepaid card,
  managed by financial             money order
  institutions or nonfinancial
  payment service providers
  (but may have pass-through
  deposit insurance for
  consumers)
  Savings accounts owned by        Checks, debit card, OBBP, BANP
  consumers ("nontransactions"
  accounts in the non-M1 part
  of M2 with direct payment
  capability)
Liabilities (credit)
  Revolving credit                 Credit card
  Nonrevolving credit              Charge card, text/SMS

OBBP. online banking bill payments; BANP,
bank account number payments; SMS, short message service.

Source: Authors' analysis and Greene,
Schuh, and Stavins (2016).

TABLE 9
DCPC Statement of Account Flows, October 2012

                                     Flows Associated
                                       with Accounts

                                 Currency    DDA     NFDA

A. Production (inflows)            388      5,379     NA
B. Consumption and                -1,038    -4,422   -58
  investment (outflows)
  B.1 Consumption                 -1,038    -4,422   -58
  expenditure
  B.2 Capital (durable              NA        NA      NA
  goods) expenditure
C. Financing                       -91       -536     -1
  C.1 Deposits (inflows)           498       564      20
    From currency                   --       564      15
    From demand deposits           455        --      2
    From nonfinancial               21        NA      --
    deposit accounts
    From foreign currency           0         NA      NA
    From long-term                  NA        NA      NA
    financial assets
    From revolving accounts         22        NA      3
    From other debt                 NA        NA      NA
    Addendum: Total                886      5,943     20
    deposits (inflows)
  C.2 Withdrawals                  -589     -1,100   -21
  (outflows)
    To currency                     --       -455    -21
    To demand deposits             -564       --      NA
    To nonfinancial deposit        -15        -2      --
    accounts
    To foreign currency             -2        NA      NA
    To long-term assets             NA        NA      NA
    To revolving accounts           NA        NA      NA
    To other debt                   -8       -643     0
    Addendum: Total               -1,627    -5,522   -79
    withdrawals (outflows)
D. Change in account               -741      421     -59
  balance (from Statement
  of Account Flows)
E. Change in account               164        NA      NA
  balance (from Balance
  Sheets)
E Flow error                       905        NA      NA
G. Error (% lagged account         135%       NA      NA
  balance)

                                      Flows Associated
                                        with Accounts

                                 Foreign             Revolving
                                 Currency    LTFA      Debt

A. Production (inflows)             NA        NA        NA
B. Consumption and                  NA        --      -1,249
  investment (outflows)
  B.1 Consumption                   NA        --      -1,249
  expenditure
  B.2 Capital (durable              NA        --        NA
  goods) expenditure
C. Financing                        2         NA        -43
  C.1 Deposits (inflows)            2         NA        NA
    From currency                   2         NA        NA
    From demand deposits            NA        NA        NA
    From nonfinancial               NA        NA        NA
    deposit accounts
    From foreign currency           --        NA        NA
    From long-term                  NA        --        NA
    financial assets
    From revolving accounts         NA        NA        --
    From other debt                 NA        NA        NA
    Addendum: Total                 2         NA        NA
    deposits (inflows)
  C.2 Withdrawals                   0         NA        -43
  (outflows)
    To currency                     0         NA        -22
    To demand deposits              NA        NA        NA
    To nonfinancial deposit         NA        NA        -3
    accounts
    To foreign currency             --        NA        NA
    To long-term assets             NA        --        NA
    To revolving accounts           NA        NA        --
    To other debt                   NA        NA        -18
    Addendum: Total                 NA        NA      -1,292
    withdrawals (outflows)
D. Change in account                2         NA      -1,292
  balance (from Statement
  of Account Flows)
E. Change in account                NA      -4,501     -673
  balance (from Balance
  Sheets)
E Flow error                        NA        NA       -619
G. Error (% lagged account          NA        NA        92%
  balance)

                                 Flows Associated
                                   with Accounts

                                 Other
                                  Debt      All

A. Production (inflows)            NA      5,767
B. Consumption and                 NA      -6,771
  investment (outflows)
  B.1 Consumption                  NA      -6.771
  expenditure
  B.2 Capital (durable             NA        NA
  goods) expenditure
C. Financing                      669        0
  C.1 Deposits (inflows)          669      1,753
    From currency                  8        589
    From demand deposits          643      1,100
    From nonfinancial              0         21
    deposit accounts
    From foreign currency          NA        0
    From long-term                 NA        0
    financial assets
    From revolving accounts        18        43
    From other debt                --        0
    Addendum: Total               669      7,520
    deposits (inflows)
  C.2 Withdrawals                  NA      -1,753
  (outflows)
    To currency                    NA       -498
    To demand deposits             NA       -564
    To nonfinancial deposit        NA       -20
    accounts
    To foreign currency            NA        -2
    To long-term assets            NA        0
    To revolving accounts          NA        0
    To other debt                  --       -669
    Addendum: Total                NA      -8,524
    withdrawals (outflows)
D. Change in account              669      -1,004
  balance (from Statement
  of Account Flows)
E. Change in account             9,489     -8,816
  balance (from Balance
  Sheets)
E Flow error                     -8,820    7,812
G. Error (% lagged account        93%       -89%
  balance)

Source: 2012 DCPC and authors' calculations.