Document Sample
Preliminary Powered By Docstoc
                                                                         Comments welcome

    What do banks do? Measuring traditional and non-
                traditional bank output

                     Robert Inklaar and J. Christina Wang*
                                      October 2006

Abstract: Most analysis of bank productivity and efficiency relies on ad-hoc measures of
traditional and non-traditional output even while the theory of banks allows for more
consistent measurement. In this paper we analyze and measure the output of U.S.
commercial banks from the point of view of banks as reducers of information
asymmetries. This proves to be a fruitful framework for dealing with traditional lending
and deposit taking as well as non-traditional activities, such as loan securitization.
Industry output growth according to this framework is noticeably different when
compared to U.S. statistical series and to methods used by bank efficiency researchers.
We also suggest ways of incorporating our measures in the analysis of bank performance.

JEL Classifications: E01, E44, O47

 Inklaar: University of Groningen and The Conference Board;
Wang: Research Department, Federal Reserve Bank of Boston;
We would like to thank Chris Kask of the BLS for providing us with the data and description of
the BLS output statistics for commercial banks, and Susanto Basu, Barry Bosworth, Erwin
Diewert, John Fernald, Alice Nakamura, Marshall Reinsdorf, Paul Schreyer, Kevin Stiroh,
Marcel Timmer, Jack Triplett, Frank Wyckoff and participants at the NBER/CRIW Summer
Institute 2006 and seminars at the Federal Reserve Bank of San Francisco and University of
Groningen for useful comments and suggestions. The views expressed in this paper are solely
those of the authors and do not necessarily reflect official positions of the Federal Reserve Bank
of Boston or the Federal Reserve System.

The measurement of bank output has long been a contentious topic, focusing on issues
like the role of deposits and the rise of non-traditional bank activities such as
securitization and derivatives.1 Statistical offices have also been interested in the topic,
but have generally chosen different approaches to the same problem.2 Although
agreement on the right output measures is still lacking, a general finding is that
measurement choices matter a great deal.3 From a methodological point of view, statistics
will not help in choosing between different approaches, so we rely on a coherent
theoretical model of bank production to derive our measure of bank output.
        In this paper, we analyze both types of bank activities with a particular focus on
non-traditional activities and provide new estimates of real output growth in U.S.
commercial banking. We compare three different measures of industry output, one
following current BLS procedure, one akin to methods used by many bank efficiency
researchers and a series based on activity counts for traditional activities and careful
deflation of fees & commissions. The three series imply very different growth of output,
and hence productivity, between 1987 and 2004 so a choice has to be made which is a
better representation of the financial services provided by the U.S. commercial banking
        The answer to this question depends on the conceptual framework of bank
production that is espoused. In the intermediation approach of Sealey and Lindley (1977),
banks are seen as producing loans and other interest-earning assets. In this framework,
there is a case to be made for treating off-balance-sheet (OBS) items symmetrically and

  See Berger and Humphrey (1997) for a general survey of bank efficiency measurement and James (1988),
Hunter, Timme and Yang (1990), Mester (1992), Jagtiani, Nathan and Sick (1995), Rogers (1998), Stiroh
(2000) and Clark and Siems (2002) for academic research on this topic.
  See e.g. Brand and Duke (1982) for the approach taken by the U.S. Bureau of Labor Statistics (BLS) and
Fixler and Reinsdorf (2006) for recent research by the U.S. Bureau of Economic Analysis (BEA).
  Even this consensus is fairly thin in places. For example, Stiroh (2000, p. 1703) finds that “efficiency
estimates are particularly sensitive to the output specification and failure to account for non-traditional
activities like off-balance sheet (OBS) items leads profit efficiency, but not cost efficiency, to be
understated for the largest [bank holding companies],” while Clark and Siems (2002, p. 987) find that “cost
X-efficiency estimates increase with the inclusion of the OBS measure [while] profit X-efficiency estimates
are largely unaffected.”

including the credit-equivalent amount of these obligations as an additional output.4 The
model of Basu, Fernald and Wang (2004) implies different prescriptions about the
treatment of on-balance and off-balance-sheet activities. In this model, the core activities
of banks are the reduction of information asymmetries and the provision of transaction
services to depositors. The transfer of funds from lenders to borrowers is only incidental
to these services. This implies that it is the services that should be valued and priced,
instead of using the accompanying loan and deposit balances as output measures. This
also provides a consistent framework for dealing with non-traditional activities, such as
loan servicing. While with loan securitization, the underlying balance and interest flows
have been repackaged and sold, the information services (screening, monitoring) are the
same.5 In some studies, net non-interest income is included as one of the outputs, but, to
our knowledge, no study makes allowances for the variety of services included in this
measure and the potentially different price development of these services.6
        We argue that our preferred industry output series gives a more insightful picture
of the production of financial services by U.S. commercial banks than the other two
series we examine. First of all, the BLS index ignores up to a quarter of industry output
by measuring only traditional activities, while our series suffers from no such omission.
Furthermore, our series is consistent with the recent theoretical framework of Basu et al.
(2004) and, more in general, with the literature arguing that the raison d’être of banks is
to reduce information asymmetries.7 In this, our series is to be preferred compared to the
somewhat ad-hoc choices made in the bank efficiency literature. Although our preferred
method cannot be applied in full at the individual bank level, we suggest a method for
incorporating a number of features from our approach into bank-level analyses.
Following this approach should provide a more accurate assessment of individual bank

  This measure tries to approximate the amount of on-balance sheet assets that would result in comparative
risk exposure for the bank. However, this need not bear a direct relationship to cost or revenue-generating
activities such as loan servicing. An alternative is the asset equivalent measure suggested by Boyd and
Gertler (1994).
  This was also noted by Mester (1992).
  Net non-interest income excludes service charges for deposit accounts and this measure was suggested by
Hunter, Timme and Yang (1990) and later used by, for example, Rogers (1998), Stiroh (2002) and Clark
and Siems (2002).
  See for example Campbell and Kracaw (1980) and Diamond (1984, 1991) for theoretical modeling along
these lines. See Mester (1992) for an empirical analysis that takes some of these considerations into

performance over time than current methods. In the next section we outline our
methodology and data sources. Section 3 compares the different output series, Section 4
discusses the implications for bank efficiency research and Section 5 concludes.

Methodology and data sources

Output at current prices
This section discusses the methodological choices and provides the outline of the data
construction, with greater detail available in the data appendix. First, we discuss output
measures at current prices, next the way in which we distinguish between price and
quantity changes for traditional and non-traditional activities and finally the method used
to aggregate changes in prices and quantities into an overall industry output series. As in
other industries, a logical starting point for measuring output is to look at total revenues.
The main data source for revenue data is the Report on Condition and Income (Call
report), in which all FDIC-insured, U.S.-chartered commercial banks provide a wealth of
information about their income, balance sheets and off-balance sheet (OBS) obligations.
        In the conceptual framework of Basu et al. (2004), banks resolve informational
asymmetries when making loans (borrower services) and provide transaction services for
deposit holders (depositor services). This last point is the source of continuing contention,
going back at least as far as Sealey and Lindley (1977) and Benston and Smith (1976).
Sealey and Lindley (1977) argue that bank transaction services yield no direct revenue
and are merely part of the cost of acquiring deposits while Benston and Smith (1976)
argue that banks produce financial services („commodities‟) for both depositors and
borrowers and are compensated for the accompanying costs. The Basu et al. (2004)
framework is a rigorous and formal expression of the latter argument. Banks acquire
funds from depositors at interest rates that are generally below the risk-free rate8 and even
though this is no „direct revenue‟, it is an economically relevant and conceptually well-
defined flow of receipts. In return for these lower interest rates, banks provide transaction

 In general, a risk adjustment has to be made for deposit accounts, but deposit insurance makes the bulk of
deposits equivalent to a risk-fee investment.

         Reasoning along similar lines, interest income is only considered a payment for
financial services insofar as the interest rate paid is higher the return earned on securities
in financial markets with comparable risk characteristics. The remainder of interest
income is excluded as it is a transfer of income from borrowers to lenders. Based on this
line of reasoning, we do not include interest income from inter-bank loans or securities as
part of output, since there are little or no financial services associated with those assets. 9
The empirical implementation of this approach is described in detail in Basu, Inklaar and
Wang (2006).
         Although the Call reports have for many years provided a considerable amount of
information about the interest income and expenses associated with a wide variety of
loans, deposits and securities, the coverage of non-traditional activities has been much
more limited. However, over the years, the Call reports have started to cover larger
number of activities and since 2001, information on 12 types of non-interest income is
collected. As a first step though, a distinction has to be made between income from fees
and commissions and income based on asset holding gains or losses. In general, only the
fees and commissions should be considered a payment for financial services, while the
latter represent transfers of income.10
         In practice, this distinction is not always straightforward to make. Some
categories, like service charges on deposit accounts, only cover fees and commissions,
just like trading revenue only covers holding gains and losses. However, venture capital
revenue includes both fees and commissions and holding gains and losses. 11 The fact that
total venture capital revenue of the commercial banking industry turned negative in 2001
and 2002 suggests that for this category, holding gains and losses dominate any fees or
         After removing all income categories predominantly associated with holding
gains or losses, we are left with seven categories that, based on the Call report
instructions and their evolution over time, cover mostly fees and commissions. Together,

   Indeed, in Europe the inter-bank rate is used by statistical agencies to approximate a risk-free, services-
free interest rate.
    See e.g. Fixler and Moulton (2001) on this distinction. As a rule of thumb, any type of income that can be
both positive and negative would not represent payment for services since negative output has no economic
    See the instructions for filling out the Call reports at

these categories account for around 90 percent of non-interest income. As described in
detail in the data appendix, most of these categories are only collected since 2001, so
other data sources and certain assumptions were used to extrapolate them back to 1987.

Quantity of loan and depositor services
The main feature of traditional bank activities these are for the most part not explicitly
priced. Instead, customers pay a higher interest rate on loans than they would have to pay
in financial markets and receive a lower rate on deposits. The higher interest rate on loans
pays for the costs associated with evaluating the creditworthiness of customers and
monitoring them for the duration of the loan. The lower rate on deposits is compensation
for the transaction services customers receive. In addition, deposit holders also have to
pay some explicit fees such as maintenance charges or fees for dropping below a
minimum balance. However, we would argue that those fees are also payments for the
transaction services provided.12
        Since these services are not explicitly priced, alternatives to collecting direct price
information need to found. The literature has suggested two alternatives, namely the
„quantity indicator‟ approach and the „deflated balances‟ approach. In the quantity
indicator approach, the number of loans and the number of deposit account transactions
are counted and used to approximate the quantity of services provided. In effect, this
assumes that a given loan or a given deposit account transaction represents the same
quantity of services over time. For the deflated balances approach, the outstanding
amounts of loans and deposits are deflated using a general price index like the CPI or the
gross domestic purchases deflator. This approach implies the assumption that a given
amount of purchasing power lent or deposited represents a constant quantity of services
over time.
        When comparing the two assumptions, neither seems perfect, but intuitively, a
constant amount of services per loan or per transaction seems more palatable. When a
customer comes to a bank for, say, a residential mortgage, it seems reasonable to assume
there is a given amount of screening, risk evaluation and administrative expenses
associated with that loan. On the other hand, if housing prices rise relative to the average

  There are also some explicit fees associated with loans, such as loan origination fees and investigation
and service charges, but these are included in the accounts as part of interest income.

price level, the size of the average mortgage is likely to increase, while the amount of
screening, etc. is unlikely to change. In a formal setting, Basu and Wang (2006) also
show that services are not proportional to (real) balances if the underlying technology for
producing the service changes over time. So while the deflated balances approach is
easiest to implement, the quantity indicator approach is likely to be a more accurate
reflection of financial services provision to lenders and borrowers.
        As mentioned before, the quantity indicator approach is far from perfect as
financial service provision is unlikely to be the same for each type of real estate loan or
stay the same over time. But however crude a measure it is, we would argue that it
captures a relevant dimension of services provision, even as it omits many features
related to the quality of services provision. On the other hand, there is no reason to
believe that using deflated balances as an output indicator will systematically give
information about financial services provision by banks. The contrast between the two
approaches becomes more obvious with a physical good analogy. Consider measuring the
price and quantity of car sales: counting the number of cars would be a reasonable first
approximation of the quantity of car sales, while there is no reason to believe that CPI-
deflated car sales will give relevant information about the quantity of cars sold. Of
course, counting the number of cars is only a first approximation as it lumps together
different types of cars, so more detailed counts would be preferable. Still, given the data
constraints we face, we choose to use detailed quantity indicators for four different types
of loans and two types of deposit accounts from the BLS to measure the growth of real
loan and deposit services.13

Quantity of fee-generating activities
In the bank efficiency literature, three output measures for non-traditional activities have
been proposed; see Clark and Siems (2002). The first of these is the credit equivalent of
OBS items, originally proposed by Boyd and Gertler (1994).14 Under the Basel capital
adequacy standards, many OBS items are assigned a risk weight so that they are
comparable to balance sheet items in terms of risk exposure. This measure mainly covers

   These are real estate, credit card, other consumer and commercial & industrial loans and demand and
time & savings deposits. We thank Chris Kask at the BLS for kindly providing these data as well as their
   We use the definitions as outlined in the data appendix of Berger and Mester (2003).

the value of letters of credit, other loan commitments and interest rate options written, but
does not include activities such as loan servicing that generate fees but do not represent a
significant risk exposure. The second is the asset equivalent measure, also proposed by
Boyd and Gertler (1994). Recognizing that the credit-equivalent measure will understate
non-traditional activities, they capitalize non-interest income since non-interest income
includes all income from OBS activities. For this capitalization, they apply the same rate
of return on assets to non-interest income. This return on assets is measured as interest
income minus interest expenses and the loan loss provision over total assets. The final
measure, proposed by Hunter, et al. (1990), is net non-interest income, or non-interest
income excluding service charges on deposit accounts.
        A curious feature of the asset equivalent measures is that effort is made to convert
an income flow into a stock, just as the credit equivalent measure is aimed at being
comparable to balance sheet items. In both cases, it implicitly introduces the assumption
that the financial services flow is somehow proportional to this stock.15 More in general,
little time is spent discussing which activities generate non-interest income and how the
price and quantity of those services should be tracked over time.16 In all fairness, the
variety of these activities, as well as the lack of detailed information from the Call reports
until recently, means that this is by no means a straightforward exercise. Nevertheless,
with the increased availability of information about non-interest income, more informed
judgments can be made about the financial services that are associated with non-interest
income and about the relevant price indices.
        We distinguish five categories of fee-generating activities, namely fiduciary
activities, investment banking, securitization activities, insurance and other activities. Our
guiding principle in coming up with output and price measures for these categories is that
there is nothing particularly special about these activities: banks provide a service and
customers pay a fee or commission in return. While the price of the services may be hard
to measure in practice, the problem is qualitatively different from that of measuring
borrower and depositor services, since those services are implicitly priced, while fees and
commissions are explicit charges.

   Presumably, these stocks are always deflated using a general price index, but this is not always obvious
from the descriptions.
   Mester (1992) is an exception as she draws the parallel between balance sheet and securitized loans.

         In the case of fiduciary activities, there are two options for arriving at a quantity
measure. First of all, the BLS constructs a measure of the number of trust accounts,
which is the only item in their output index covering fee-generating activities. Second,
there is a personal consumption expenditure (PCE) deflator that covers trust fees. In
practice though, the PCE deflator is also based on a count of the number of trust
accounts, but only for personal accounts, not corporate accounts.17 Therefore we use the
BLS quantity index, which gives a complete coverage of trust accounts.
         For two other categories, deflation is relatively straightforward, namely
investment banking and insurance. In both cases, we use the industry gross output
deflator from the BEA GDP by Industry accounts. Securitization activities fees and
commissions earned based on loans that are no longer on the bank‟s balance sheet but
which the bank still services. As discussed in Mester (1992), banks will chose to
securitize some loans while keeping others on their balance sheets so they will not be
perfect substitutes. However, as a first approximation, it seems reasonable to assume that
the financial services provided should be relatively similar to loans of the same type that
remain on the balance sheet.18 We therefore use the implicit price based on BLS quantity
indicators to come up with an implicit price index for securitization activities.19
         The final category of fee-generating activities is a residual of other activities,
labeled „other noninterest income‟ in the Call reports. According to the Call report
instructions, this should cover some 25 types of fees and commissions as well as any
other fees and commission. Some of the examples mentioned are credit card fees,
issuance of commercial letters of credit and certain types of loan commitment fees.
Furthermore, even if more detail was available about this category, it is doubtful than any
suitable price index would be available.20 For lack of a more appropriate price index, the

   Since December 2003, the BLS collects data on the price of trusts services directly as part of its PPI
program. Given the short history, this data cannot be used yet.
   For example, it seems quite reasonable to assume that a bank first has to spend a fixed amount of effort to
ascertain the risk of a loan and only later, on the basis of this rating, decides to sell the loan or keep it in the
balance sheet.
   There are two income categories that fall under this heading, namely net servicing fees and net
securitization income, each covering part of the securitization process. Since we have no further price or
quantity information, we combine these categories and apply the same price index to overall securitization
   On the Call reports, further details should only be provided if the category in question represents more
than 1 percent of total revenue, which is rarely the case. In the Economic Census, more revenue categories
are collected, but in the 2002 Census, the product categories for fee-generating activities do not overlap the

PCE deflator for service charges on deposit accounts is used, the solution also chosen by
the BEA. This is clearly the weakest point as this category accounts for a substantial
share of output (see the next section) and includes revenues associated with a number of
OBS items. Still, the fact that improvements are still possible should not distract from
what is, in our view, progress in most categories. Moreover, by casting the problem in
terms familiar from price measurement in other industries (i.e. identifying and measuring
the prices of explicitly priced goods or services), the call for improvements in
measurement by statistical offices becomes more straightforward.

To come up with an overall measure of bank output, the real output series need to be
aggregated in some way. Here we follow a non-parametric approach, which means that
given a set of weights and an index number formula, an overall output series can be
calculated. In much of the bank efficiency literature, the translog function is used in
estimating cost and profit functions. The BLS also frequently relies on the Törnqvist
index, an exact index for the translog production function, as its index of choice for
aggregation.21 We therefore use a Törnqvist index to combine our output series:

          ln q t ,t 1  
                              1 t
                                wi  wit 1  ln qit ,t 1   wi  ln qit ,t 1 .                  (1)
                        i                                    i

        Equation (1) states that the percentage growth of total output q from period t-1 to t
is equal to the weighted average growth of each of the components of output, qi. The
weight used is the two-period average share w of component i in the total. In general, w is
calculated as the share in total output, which is what we use in our preferred output series.
Note that, as argued above, we consider service charges on deposit accounts to be a
payment for transaction services, so the imputed output related to deposit accounts is
combined with these service charges to determine overall deposit-related output.
        In the case of commercial banking, the BLS does not use output shares but instead
employment requirements shares based on the Functional Cost Analysis (FCA). The FCA
used to be held once every five years, but has been discontinued since the 1997 survey.

Call categories in some cases and moreover, the „other services‟ category accounts for almost half of fee
and commission revenue.
   See e.g. Diewert (1976) on the Törnqvist index.

Furthermore, the FCA was a voluntary survey in which mostly smaller banks
participated, so the shares are in general not representative of the industry as a whole. 22 In
the next section, we will examine the impact of using employment shares rather than
output shares to come up with output series for loans and depositor services.23 In the bank
efficiency literature, the different outputs are not aggregated into a single output index,
but instead included separately in a cost or profit function. As our focus is on the quantity
measures and less on the aggregation, we use the same output weights for the „bank
efficiency‟ output measure as for our preferred measure when aggregating traditional and
non-traditional activities.

As discussed in the introduction, the activities of commercial banks have changed
considerably in the past 15-20 years. Table 1 illustrates this by comparing the
composition of bank output at current prices in 1990 and 2004.24 The table shows that
about half of the growth in current output over this period can be attributed to growth in
fees and commissions and the share of this category has increased from 27 to 41 percent.
On the other hand, the share of deposits decreased by 18 percentage points. As the
breakdown of fees and commissions demonstrates, part of the increased importance of
this category can be traced to non-bank activities like investment banking and insurance.
However, even in 2004 these activities only make up a modest portion of bank output
with a combined share of less than 5 percent.
        A much more influential shift is the increase in securitization revenue, which has
grown from just over one percent to more than 11 percent of output.25 This sheds a
somewhat different light on the label „non-traditional bank activities‟ since such a large

   See Ors (2004).
   The BLS uses the same shares for five years before switching to new shares, thereby creating a
Laspeyres index. Here we interpolate the shares linearly between the FCA years and apply the Törnqvist
index from (1). This leads to some differences compared to the BLS bank output index. Moreover, we only
estimate output growth for U.S.-chartered commercial banks and not for branches of foreign banks, since
data availability fall short for those branches. Together these factors drive a wedge between our BLS-based
numbers and the published output index.
   Our data series start in 1987, but output related to loans is uncommonly low in 1987-1989 due to
relatively high risk premia, so 1990 is shown to give an indication of the broader trends over this period.
   When looking at bank balance sheets, securitized loans have increased from about 10 percent of total
loans (i.e. balance sheet and securitized loans) to more than 40 percent.

part of those revenues are still related to loan provision. The only difference in an
economic sense is that in the case of securitization activities, the underlying principal is
not kept on the balance sheet but resold. Despite their growing importance, bank
securitization activities are ignored by both the BLS and, as discussed above, mostly
omitted by bank efficiency researchers when using the credit equivalent of OBS items.
         Table 2 compares various measures of borrower and depositor services for the
period 1987-1995 and 1995-2004. The first line in each panel shows total balances
deflated using the price index of gross domestic purchases.26 The second line in each
panel is based on the detailed loan and transaction counts from the BLS and uses
employment weights to combine these into indices, while the third line uses output
weights for aggregation. There are obvious differences between the deflated loans and
deposits and activity counts measures, in particular in the case of deposits. While
balances rose rapidly after 1995, the number of transactions declined over the same
period. In general, there is no relationship between overall inflation and changes in the
balance-to-count ratio,27 with the correlation varying between -0.6 and +0.3. On the other
hand in the case of residential real estate loans, there is a clear correlation between the
average mortgage size and housing prices.28 Although no similar correlation could be
found for other activities, it does suggest the activity counts measure a relevant part of
financial services provision.
         The choice of weights also has an impact, as output weighted loan counts show
higher growth and deposit transaction counts show consistently lower growth than the
employment weighted counts. This could point to noticeable labour productivity
differences between the different activities. Alternatively, it could mean that the
employment weights from the FCA are not representative, either because of an

   This index shows a very similar pattern over time as the GDP deflator. In Panel A, only loans are
included for comparability to the other measures, but average growth would be similar if all interest-
earning assets were included. Shown in the table is the sum of loans and the sum of deposits, deflated using
this price index. The growth patterns are comparable if output weights would be used to aggregate detailed
loan and deposit categories.
   The change in the balance-to-count ratio measures changes in the average loan size or average number of
transactions per dollar of deposits. Appendix Table A.5 reports the indices of these ratios for each activity
covered by the BLS.
   Residential real estate loans are a component of overall real estate loans and average mortgage size is
published by the Federal Housing Finance Board. Comparing the average loan size and the housing price
index from OFHEO shows a positive correlation of 0.8.

unrepresentative sample or because of changes in weights over time. Either way, the
current BLS weighting procedure seems problematic.
         Table 3 compares the different measures of real growth of fee-generating
activities. First, the three proposed measures from the bank efficiency literature are
shown, namely the credit equivalent of OBS items, the asset equivalent and net non-
interest income. All three measures at current prices are deflated using the price index for
gross domestic purchases. Each measure shows a considerably different growth pattern
over time, with the credit equivalent measure showing high but declining growth, the
asset equivalent measure showing increasing growth and net non-interest income
showing more moderate and stable growth. It is useful to recall that the asset equivalent
measure is simply non-interest income divided by the financial return on assets,29 so the
difference between the two measures reflects the rising return on assets before 1995 and
the declining return after 1995. However, there is little basis for assuming that the diverse
range of fee-generating activities would have exactly mimicked this time pattern of
returns. In general, the large differences between the bank efficiency measures make a
more reasoned and defensible measure of large importance. Furthermore, as Stiroh
(2000) found, individual bank efficiency estimates are (predictably) sensitive to the
output measure chosen.
         The BLS measure of fee-generating activities only includes a count of the number
of trust accounts and between 1987 and 2004, growth in this category has been very low.
However, as Table 1 showed, fiduciary activities make up only a modest and shrinking
share of fee-generating activities, declining from about a quarter to 17 percent of output.
The bottom part of Table 3 shows the results from our more comprehensive measure,
which covers all of net non-interest income. The „total fee-generating activities‟ line
shows that moderate and increasing real growth of fee-generating activities. Note that the
difference between the „net non-interest income‟ line and the „total fee-generating
activities‟ line is in the price indices used to deflate the same output at current prices. The
reason that fee-generating activities show slower growth is mainly because of faster
growth in prices for „other activities‟. As Table 1 showed, this category accounted for

  Also recall that this is measured as net interest income (after subtracting loan loss provisions) over total

two-thirds of fee-generating output in 1990 and in 2004 it still makes up nearly half of
output. As discussed in the previous section, this is a residual category including many
different types of services, from safety deposit boxes to loan commitment fees and our
use of the PCE deflator for service charges on deposit accounts may or may not be
appropriate. However, even if we were to use the price index for gross domestic
purchases, the pattern of moderate but increasing real output growth would be
unaffected.30 This is mainly because of the increased importance of securitization
activities, which accounted for about two-thirds of real growth in fee-generating
activities. Investment banking has also contributed substantially to overall growth.
         Although it is already obvious that the three approaches to measuring bank output
lead to very different outcomes for both traditional and non-traditional activities, it is
useful to compare overall bank output growth. Obviously, this overall measure will
reflect the differences found at a detailed level. To recap the main ones, loan and deposit
transaction counts are very different from deflated balances, output weights yield
different results than employment weights and detailed deflation leads to a different
growth pattern for fee-generating activities than the alternatives.
         Aggregating over the different activities is straightforward in the case of the BLS
index and our preferred measure, since the former is based on employment weights,
while our measure is based on a consistent set of output values, prices and quantities.
However, in the bank efficiency literature, the different output measures are used in
estimating cost and profit functions and an overall output measure is generally not an
output of such an analysis.31 What we do here is combine the results from Tables 2 and 3
using our output weights to construct a number of industry output growth measures that
correspond to different assumptions made in the bank efficiency literature. As discussed
in the previous section, our aggregation methodology is consistent with a translog
production function, so the assumptions are comparable to assuming a translog cost or
profit function.32

   The average growth for total fee-generating activities before and after 1995 would be 5.4 and 7.5 percent,
   Some studies do analyze overall productivity growth in the industry; see e.g. Stiroh (2000) for an
overview of approaches and results.
   Moreover, given the detailed results presented in the Appendix tables and the readily available industry
measures from the Call reports, any number of alternatives can easily be calculated.

        Table 4 reports the results from this exercise. We show six different series that are
calculated in the spirit of the bank efficiency literature. The first three assume that both
loans and deposits are output, so the growth figures from Table 2, panel C are used that
show deflated loans and deposits. The remaining three assume that depositor services are
not an output, so the deflated loans from Table 2, panel A are used. Apart from the issue
of whether deposits are output or inputs, the output measure of OBS items is also
examined, using the growth rates from Table 3. The BLS measure uses the activity counts
based on employment weights from Table 2 and fiduciary activities from Table 3.
Finally, our preferred measure is based on the activity counts with output weights from
Table 2 and total fee-generating activities from Table 3.
        The main finding from Table 4 is that the assumptions made about bank output
matter a great deal for evaluating the performance of the U.S. commercial banking
industry. Growth over the period 1987-1995 varies between 1.2 and 12.8 percent on
average per year, while growth between 1995 and 2004 varies between 0.9 and 8.7
percent. Furthermore, some measures show improving growth after 1995, other show
declining growth and there is even a measure showing stable growth. With such
differences, an agnostic view of what best represents financial services provision by
commercial banks does not seem like an appealing option. We would therefore argue for
adopting current theoretical and measurement insights and relying on our preferred
measure of bank output. This measure shows modest growth of 1.9 percent on average
per year before 1995, accelerating to 2.5 percent after 1995.

A measurement guide for bank efficiency research
Our measurement of industry output for U.S. commercial banks cannot directly be
applied to individual banks. However, there are a number of features which can be
incorporated in a relatively straightforward fashion. First of all, the data on average loan
size and deposit transactions are only available industry-wide. However, it would be
fairly straightforward to use these industry trends to „deflate‟ individual bank balances
(see Appendix Table A.5).33 While this is not ideal, it is common practice in micro-level

  For even further refinement in the area of residential mortgages, one could even use local housing price
indices in combination with data on the location of deposit holders to deflate these loans.

research to use industry prices to deflate sales of firms or establishments and moreover, it
would be a definite improvement over current practice of using a general price index.
       The increased information on non-traditional activities since 2001 also facilitates
improvements in measurement in that area. First of all, information about trust accounts
(in Call report schedule RC-T) allows for a bank-specific count of the number of
accounts, while for earlier years the (implicit) price index of the industry can be used.
Revenue from these activities is available for many years. Second, Call report schedule
RC-S provides information on the balances of securitized loans by type. The appendix
details how these balances can be extrapolated for years before 2001 using industry-wide
trends and how securitization revenue can be extrapolated using the trend in servicing
assets. Bank-specific information on trends in investment banking and insurance revenue
is hard to come by, so there would be seem to be few alternatives but to use industry-
wide trends in revenue to extrapolate to years before 2001. Given that these revenues are
for the most part a small share of revenue, especially before 2001, this should have a
relatively impact.

In this paper we argue that the current approaches to measuring the output of banks used
by bank efficiency researcher and statistical offices are due for an overhaul. Recent
theoretical advances provide a coherent framework for accounting for traditional lending
and deposit-taking activities and improvements in data collection in the U.S. allow for a
more accurate assessment of the role of non-traditional activities, such as investment
banking and loan securitization. We construct a new output series for the U.S.
commercial banking industries that covers the period 1987 to 2004 and compare this
series to alternatives based on the bank efficiency literature and the BLS methodology.
Our starting point is that the primary role of banks is to reduce information asymmetries
when making loans and perform transaction services for deposit holders (see Basu et al.
2004). This allows us to distinguish between interest transfers between (ultimate)
borrowers and lenders on the one hand and payments for financial services on the other
hand. Second, when it comes to activities that generate fees and commissions, the most
straightforward output measure (at current prices) is total revenue.

       To arrive at series of real output growth, we use information about the number of
loans and deposit transactions from the BLS as quantity indicators for traditional bank
activities. We argue that this is a definite improvement over the common practice of
deflating loan and deposit balances using a general price index, even though further
improvements would still be possible and preferable. For non-traditional activities, we
use a combination of price and quantity information to arrive at output series for five
distinct activities, namely fiduciary activities, investment banking, insurance, loan
securitization and other activities. We contrast our resulting preferred output measure to a
series of measures used in the bank efficiency literature and the output measure used the
BLS and find that there are important differences, both in the average growth over the
entire period and the pattern of growth before and after 1995. Given that 1995 is most
commonly used as the starting point of the U.S. productivity growth acceleration, it is of
considerable interest to know whether output (and hence productivity) growth increased
or decreased after 1995 compared to the earlier period.
       We know of no statistical measure that would allow us to prefer one measure of
output over the other. However, we would argue that unlike other measures, our preferred
measure is consistent with a coherent theoretical framework. So, for example, the BLS
measure ignores up to a quarter of total bank output and the bank efficiency literature
implicitly equates stocks of balance sheet and off-balance-sheet items with financial
services provision. In coming up with a preferred output measure for U.S. commercial
banks, we by no means would argue that this is as good as it is going to get, only that it
exploits all available data sources and current theoretical insights. Adopting this measure,
both in statistical practice and current research would avoid confusion over differences in
output measures and furthermore, focus the attention on areas where measurement could
and should still be further improved.

Basu, Susanto and J. Christina Wang (2006), “Technological Progress, "Money" in the
      Utility Function, and the „User Cost of Money‟ ” paper presented at the
      NBER/CRIW Summer Institute 2006, downloadable at
Basu, Susanto, John Fernald and J. Christina Wang (2004), “A General-Equilibrium
      Asset-Pricing Approach to the Measurement of Nominal and Real Bank Output”
      Federal Reserve Bank of Boston Working Papers, no. 04-7.
Basu, Susanto, Robert Inklaar and J. Christina Wang (2006), “The Value of Risk:
      Measuring the Services of U.S. Commercial Banks” paper presented at the
      NBER/CRIW Summer Institute 2006, downloadable at
Benston, George J. and Clifford W. Smith (1976), “A Transactions Cost Approach to the
      Theory of Financial Intermediation” Journal of Finance, vol. 31 no. 2, pp. 215-31.
Berger, Allen N. and David B. Humphrey (1997), “Efficiency of financial institutions:
      International survey and directions for future research” European Journal of
      Operational Research, vol. 98, pp. 175-212.
Berger, Allen N. and Loretta J. Mester (2003), “Explaining the dramatic changes in
      performance of US banks: technological change, deregulation, and dynamic
      changes in competition” Journal of Financial Intermediation, vol. 12, pp. 57-95.
Boyd, John H. and Mark Gertler (1994), “Are Bank Dead? Or Are The Reports Greatly
      Exaggerated?” Federal Reserve Bank of Minneapolis Quarterly Review, vol. 18 no.
      3, pp. 2-23.
Brand, Horst and John Duke (1982), “Productivity in commercial banking: computers
      spur the advance” Monthly Labor Review, December, pp. 19-27.
Campbell, Tim S. and William A. Kracaw (1980), “Information Production, Market
      Signalling, and the Theory of Financial Intermediation” Journal of Finance, vol. 35
      no. 4, pp. 863-82.
Clark, Jeffrey A. and Thomas F. Siems (2002), “X-Efficiency in Banking: Looking
      beyond the Balance Sheet” Journal of Money, Credit, and Banking, vol. 34 no. 4,
      pp. 987-1013.
Diamond, Douglas W. (1984), “Financial Intermediation and Delegated Monitoring”
      Review of Economic Studies, vol. 51 no. 3, pp. 393-414.
Diamond, Douglas W. (1991), “Monitoring and Reputation: The Choice between Bank
      Loans and Privately Placed Debt” Journal of Political Economy, vol. 99 no. 4, pp.

Diewert, W. Erwin (1976), “Exact and Superlative Index Numbers” Journal of
      Econometrics, vol. 4 no. 2, pp. 115-45.
Fixler, Dennis J. and Brent R. Moulton (2001), “Comments on the treatment of holding
      gains and losses in the national accounts” paper for the OECD Meeting of National
      Acconts Experts, downloadable at
Fixler, Dennis J. and Marshall Reinsdorf (2006) “Computing Real Bank Services” paper
      presented at the NBER/CRIW Summer Institute 2006, downloadable at
Hunter, William C., Stephen G. Timme and Won Keun Yang (1990), “An Examination
      of Cost Subadditivity and Multiproduct Production in Large U.S. Banks” Journal of
      Money, Credit, and Banking, vol. 22 no. 4, pp. 504-25.
Jagtiani, Julapa, Alli Nathan and Gordon Sick (1995), “Scale economies and cost
      complementarities in commercial banks: On-and off-balance-sheet activities”
      Journal of Banking & Finance, vol. 19, pp. 1175-89.
James, Cristopher (1988), “The use of loan sales and standby letters of credit by
      commercial banks” Journal of Monetary Economics, vol. 22, pp. 395-422.
Mester Loretta J. (1992), “Traditional and nontraditional banking: An information-
      theoretic approach” Journal of Banking & Finance, vol. 16, pp. 545-66.
Ors, Evren (2004), “Postmortem on the Federal Reserve‟s Functional Cost Analysis
      Program: how useful was the FCA?” Review of Financial Economics, vol. 13, pp.
Rogers, Kevin E. (1998), “Nontraditional activities and the efficiency of US commercial
      banks” Journal of Banking & Finance, vol. 22, pp. 467-82.
Sealey, Calvin W. and James T. Lindley (1977), “Inputs, outputs, and a theory of
      production and cost at depository financial institutions” Journal of Finance, vol. 32
      no. 2, pp. 1251-66.
Stiroh, Kevin J. (2000) “How did bank holding companies prosper in the 1990s?”
      Journal of Banking & Finance, vol. 24, pp. 1703-45.

Table 1, Output of U.S. commercial banks at current prices, 1990 and 2004
                                   Billions of dollars                 Share in total
                                   1990         2004                  1990       2004
Total                              123.1        322.0                 100.0      100.0
 Deposits                           65.5        115.5                 53.2        35.9
 Loans                              24.4         74.6                 19.8        23.2
 Fees & commissions                 33.3        131.9                 27.0        41.0
 of which:
   Fiduciary activities             7.9           22.6                 6.4        7.0
   Investment banking               0.8            9.7                 0.7        3.0
   Securitization activities        1.5           36.5                 1.2       11.4
   Insurance                        0.1            4.2                 0.1        1.3
   Other activities                 22.9          58.8                18.6       18.3
Notes: Output associated with deposits and loans is based on the interest margins of Basu
et al. (2006). See the data appendix for details on other items.

Table 2, Average annual real growth of borrower and depositor services of U.S.
commercial banks, 1987-2004
                                                           1987-1995         1995-2004
A: Real growth of borrower services
Deflated loans                                                 1.8               5.3
Loan counts (employment weights, BLS)                          1.7               3.7
Loan counts (output weights, preferred)                        5.9               4.3
B: Real growth of depositor services
Deflated deposits                                               0.1               4.7
Transaction counts (employment weights, BLS)                    1.6              -1.2
Transaction counts (output weights, preferred)                 -0.3              -2.3
C: Real growth of depositor and borrower services
Deflated loans and deposits                                    0.8               5.0
Activity counts (employment weights, BLS)                      1.6               1.0
Activity counts (output weights, preferred)                    1.2               0.4
Notes: Deflated loans and deposits is the growth of year-average balances from the Call
reports, deflated using the price indes of gross domestic purchases from the U.S. NIPA.
Loan counts, transaction counts and employment are provided by Chris Kask of the BLS
and output weights are based on the same data as Table 1.

Table 3, Average annual real growth of fee-generating activities of U.S. commercial
banks, 1987-2004
                                                       1987-1995                1995-2004
Bank efficiency measures
Credit equivalent                                          20.0                     11.0
Asset equivalent                                            0.9                      9.5
Net non-interest income                                     7.0                      7.3
BLS measure
Fiduciary activities                                        0.4                     0.1
Preferred measure
Total fee-generating activities                             3.3                     6.6
Contributions from:
 Fiduciary activities                                       0.1                     0.0
 Investment banking                                         0.6                     1.1
 Securitization activities                                  0.6                     4.1
 Insurance                                                  0.1                     0.3
 Other activities                                           2.0                     1.0
Notes: Credit equivalent measure is the risk-weighted sum of OBS items, based on the definitions
in the data appendix of Berger and Mester (2003). Fiduciary activities is based on data provided
by Chris Kask of the BLS. Methods used in deriving other fee-generating activities are detailed in
the main text and appendix. Contributions to total fee-generating activities are calculated by
multiplying annual growth rates by the two-period average output share (see equation (1)).

Table 4, Average annual growth of U.S. commercial bank output, 1987-2004
                                                            1987-1995              1995-2004
Bank efficiency measures
Loans, deposits & OBS items as output
Credit equivalent                                                 6.2                  7.2
Asset equivalent                                                  1.4                  4.2
Net non-interest income                                           2.9                  3.0
Only loans & OBS items as output
Credit equivalent                                                 12.8                 8.7
Asset equivalent                                                   1.2                 7.2
Net non-interest income                                            6.6                 6.1

BLS measure                                                       1.5                  0.9

Preferred measure                                                 1.9                  2.5
Notes: Bank efficiency measures combine different series from Tables 2 and 3. The items under
'loans, deposits & OBS items as output' use the growth of deflated loans and deposits (Table 2,
panel C) and each of the three OBS items from Table 3 and combines them using output shares
based on the same data as Table 1 for loans and deposits on the one hand and fee-generating
activities on the other hand. The items under 'Only loans & OBS items as output' use the growth of
deflated loans from Table 2, panel A. BLS measure is calculated using employment weights and the
relevant items from Table 2 and 3. Preferred measure is calculated using output weights and items
from Table 2 and 3.

Data appendix
This appendix covers in detail the different sources that were used as well as some of the
choices made in situations were data was not available. Basu, Inklaar and Wang (2006)
describes the data construction on the imputed value (at current prices) of deposit and
loan services in detail, so here we deal with constructing time series on the different types
of fee-generating services and other non-interest income as well as the price and quantity
data. Furthermore, detailed tables with the Call report items used and the underlying
prices and values are also shown.

The value of fee-generating services at current prices
As described in the main text, we distinguish seven different fee-generating services.
Excluded from overall non-interest income are trading account revenue, venture capital
revenue and gains and losses on the sale of loans, other real estate and other assets.
Appendix Table A.1 gives an overview of the Call report items used, while Table A.2
shows the time series for each of these activities. For the period from 2001 onwards, the
Call reports provide information on each of these categories. For the period before 2001,
the Call reports provide information on two of these, namely fiduciary activities and
service charges on deposit accounts. As argued in the main text, we consider service
charges on deposit accounts to be payments for transaction services, so we add the
service charges to the imputed output associated with deposits.34
        In the case of investment banking and insurance, the only data source with
information about the importance of these activities was the Economic Census. For 1992,
1997 and 2002, a detailed breakdown of commercial bank revenue is available. The only
drawbacks are that there is no information about 1987 or about insurance revenue in
2002. However, for investment banking, the 2002 Census provides a useful verification
compared to the Call reports: the Call reports show investment banking revenue of
$9.2bln in that year, while the Census shows $8.5bln. Since the definitions of revenue
sources do not match, this can be considered a fairly close correspondence. Furthermore,
the $8.5bln from the Census only represents fees and commissions related to investment

  Since we have two types of deposits, namely demand deposits and time & savings deposits, we distribute
the service charges in proportion to the outstanding balance of both types of deposit accounts.

banking activities, so insofar as the discrepancy between the two sources can be
attributed to the omission of any holding gains and losses from the Census definition, the
Call report figure represents predominantly fees and commissions. As a further
confirmation of the usefulness of the Census data, the revenue figures for fiduciary
activities and service charges on deposit accounts are also generally within 10 percent of
the respective Call report items.
       Since the Census is the only data source on insurance and investment banking
revenue data before 2001 that we are aware of, we use the Census figures for 1992 and
1997 and assume a constant growth of revenue for the years between observations. To
form a plausible estimate for 1987, we look at the share of investment banking and
insurance revenue in fee-generating services output, excluding fiduciary activities and
service charges on deposit accounts. In 2001, this share was 9.7 percent for investment
banking and 3.1 percent for insurance. Based on the Census, the corresponding figures in
1997 were 6 and 1.3 percent and in 1992 they were 4.4 and 0.8 percent. Given this rapid
rise during the 1990s, we assume that in 1987, the shares were respectively 2.8 and 0.4
percent, i.e. the same rise in percentage points between 1987 and 1992 as was observed
between 1992 and 1997. Obviously, without extra information, one could just as easily
make other assumptions, but given that our 1987 estimate implies only a small role for
these activities, the impact on the overall results from alternative (plausible) assumptions
would be limited.
       Two of the other categories are closely related, since they refer to net servicing
fees and net securitization income. In both cases, the revenues are related to loans that do
not appear on the bank‟s balance sheet but that were originated and/or still serviced by
the bank. So, for example, this covers the situation where a set of mortgages is pooled,
sold to Fanny Mae (FNMA), but where the bank still collects interest and repayment of
the principal. As before, estimates need to be made of revenue before 2001, but in this
case, we can make use of not just the Census, but also information in the Call reports on
securitized loans and servicing assets.
       In the case of servicing fees, the Call report instructions suggest a strong link with
servicing assets, which are part of a bank‟s intangible assets and which should reflect the
value of the future stream of servicing fees. Indeed, the annual growth of net servicing
fees between 2001 and 2005 is highly correlated with the annual growth of servicing

assets (0.90). Although this represents only a few observations, further evidence is found
in the cross-section for the different years. In 2001, the first year in which servicing fees
were included in the Call reports, the correlation across banks between servicing fees and
servicing assets was 0.5, rising to 0.7 for banks reporting positive net servicing fees, i.e.
banks with higher servicing assets generally had higher servicing fees. By 2004, this
correlation had risen to 0.8 (in both cases), suggesting that reporting consistency
improved over the years. The cross-sectional correlation between servicing fees and the
principal amount of securitized loans is much weaker, at only 0.5 in 2004 and 0 in 2001.
       Net securitization income is harder to make sense of. First of all, based on the
Call report instructions, it should include fees on securitizations, structured finance
vehicles and administrative support, but also holding gains and losses related to
securitization transactions. The time-series correlation with servicing assets is much
weaker: the overall growth between 2001 and 2005 of servicing assets is negatively
correlated with the growth in net securitization income. The cross-sectional correlations
are also weaker, varying between 0.3 in 2001 to 0.6 in 2004. However, the correlation
with servicing assets is still higher than with the principal amount of securitized loans,
which is between 0 and 0.3.
       This suggests that information on servicing assets would be most informative for
both servicing fees and securitization income in earlier years. It is therefore useful that
the Call reports provide information about mortgage servicing assets back until 1987.
While the „other servicing assets‟ category before 1992 is part of „other intangible assets‟,
they make up a substantial share of other intangible assets, so the trend in other intangible
assets can be used to come up with an estimate of total servicing assets. The amount of
securitized loans is also estimated back to 1987 for the same four categories as balance
sheet loans (real estate, credit card, other consumer and commercial & industrial loans).
From 2001 onwards, information on these categories is available from the Call reports,
while before 2001, only real estate loans are available directly. For the two consumer
loan categories, we use the trend in total outstanding private asset-backed securities
(ABS) of consumer loans from the Flow of Funds. For real estate loans before 1992, we
use the trend in total outstanding mortgage-backed securities (MBS), also from the Flow
of Funds. For other loans (including commercial & industrial loans), we use the trend in
real estate and consumer securitized loans. Since the other loans make up only 4 percent

of total securitized loans, even in 2004, the exact extrapolation method will have only a
small impact on the overall results.
       The trend in servicing assets and securitized loans are not the only pieces of
information we can use about the evolving importance of securitization revenue. The
Economic Census of 1992 and 1997 also provide information on overall securitization
revenue, although a similar category was absent in 2002. The definition is also not fully
comparable both across Census years and between the Census and Call reports. As
discussed above, the Call reports distinguish net servicing fees and net securitization
income, which includes both fees and some holding gains and losses. In 1992, the closest
Census category refers to “loan (and line of credit) servicing fees collected after
placement,” while in 1997 the definition is “loan servicing and administration fees.” This
suggests that the 1992 Census figure includes more revenue than covered under the Call
report definition, namely line of credit servicing, but also less revenue because it does not
include holding gains and losses, while it is not clear whether it includes any of the fees
included in net securitization income in the Call reports. The 1997 Census definition will
include less revenue since it does not cover holding gains and losses.
       Due to the ambiguity in the definitions and the pieces of information from
multiple sources, we developed a few different options for extrapolating securitization
revenue before 2001. Table A.3 and Figure A.1 show some of these options. The first
option shown in Table A.3 is to use the Census benchmarks in both 1992 and 1997, while
the other options use trends of servicing assets and/or securitized loans. These
extrapolations suggest that in 1997, the Census definition underestimates the Call report
definition. While this is a conceptual improvement since it removes holdings gains and
losses, one loses out in consistency over time, since we have no information to remove
holding gains and losses in the post-2001 period. In 1992, the results are more mixed,
with the trend based on servicing assets suggesting that the Census definition overstates
securitization revenue and the other two trends estimates suggesting the opposite.
       Figure A.1 also suggests some issues to take into consideration. While securitized
loans suggest a fairly monotonous rise in securitization revenue, the trend in servicing
assets suggest more variability, with in particular a decline in 2001. Note that we did not
linearly interpolate growth rates between Census benchmarks, but used the variation in
the growth servicing assets. Therefore, the general trend closely resembles the trend in

servicing assets. Based on all the evidence we collected, a clear-cut decision about which
series is best is not possible. However, we have chosen to use the series based on the
trend in servicing assets and for a few reasons. First of all, the cross-sectional correlations
for in particular 2004 suggest that servicing assets are a good predictor of servicing fees
and a reasonable predictor of securitization income. The balance of securitized loans is a
notably weaker indicator. Second, the definitional inconsistencies between Call reports
and the Census make it hard to use Census information directly with any degree of
confidence. Finally, growth between 2001 and 2005 suggests that securitization income
does not rise as monotonously as securitized loans, and that servicing assets provide a
better approximation.
       While there are uncertainties about which series of securitization revenue is the
best reflection of bank securitization revenue, we would like to stress that taking this
revenue stream into account is important, as it represents 22 percent of fee-generating
services revenue in 2004, up from only 3 percent in 1987 according to our estimate based
on the trend in servicing assets. This makes it a more important revenue category than
fiduciary activities and service charges on deposit accounts and one of the most important
sources of revenue growth of commercial banks since the late 1980s. Put in a different
light, if the securitized loans had remained on bank balance sheets, the interest income on
the four loan categories covered, at the observed average interest rate, would rise from
$250bln to $425bln. Even though we have chosen to base our estimates on the series
based on the trend in servicing assets, industry estimates based on the other options are
available upon request.
       After accounting for the trend in investment banking, insurance and securitization
revenue, the remaining category is „other non-interest income.‟ For the 2001-2004 years,
this category is directly observed, but for earlier years, it is calculated as a residual. As
discussed in the main text, this category includes a large number of potential revenue
sources and neither the Call reports nor the Census provide much guidance about the bulk
of revenue in this category. This makes this revenue estimate the least reliable. In
combination with a deflator that does not directly cover any of the activities this implies
that real growth for this category should not be used to draw far-reaching conclusions.

Price and quantity data
Most relevant information about the deflation of fee-generating activities was described
in the main text. The underlying series used for deflation are given in Tables A.4 and A.5.
Table A.4 shows the price deflators that can be used to calculating quantities of fee-
generating services in combination with the data in Table A.2. The service charges on
deposit accounts are included in both tables, even though we include the value of those
charges from Table A.2 in the imputed output for deposit accounts. In case one is
interested in only estimating the quantity of imputed depositor services, the value and
price series from Tables A.2 and A.4 can be used to „back out‟ the explicit charges.
       As discussed in the main text, the service charges price deflator is based on a PCE
deflator and this same index is used for other non-interest income. The price index for
fiduciary activities is based on the quantity count from the BLS and the value of those
services from Table A.2. For investment banking and insurance revenue, the relevant
gross output deflator from the BEA GDP by Industry accounts is used. To be precise,
these are the deflators for Securities, commodity contracts, investments (NAICS 523) and
Insurance carriers and related activities (NAICS 524).
       For net servicing fees and net securitization income, some more discussion is
needed. The same implicit price deflator is used for both series and this deflator is based
on the BLS loan quantity counts. Table A.5 shows an index of the balance-to-count ratios
for the different categories. The change ratios are calculated by taking the growth in the
relevant year-average balance of loans or deposits and subtracting the growth in the
relevant quantity count. In the case of loans, any rise in this figure implies a rise in the
average loan size. The information in Table A.5 is therefore sufficient to come up with a
quantity index for depositor and loan services.
       We also use the BLS quantity counts for securitization revenue. We assume that,
for example, a real estate loan that is on a bank‟s balance sheet require the same financial
services as a securitized real estate loan. Obviously, this need not be true in practice, but
it is the only assumption we can make without further information. This means 1) that the
balance-to-count ratios can be applied to the securitized loan balances and 2) that the
interest margins for the different loans from Basu, Inklaar and Wang (2006) can be used
to come up an imputed output estimate. The change in this output estimate for each type

of loans is calculated and the change in the balance-to-count ratio is subtracted to come
up with an implicit price index for the financial services associated with each loan type.
This implicit price index for each loan type is then aggregated using output shares to
come up with the overall implicit price index for securitized loans. This implicit price
deflator is quite volatile, with a particularly steep rise from 1987 to 1988. The „87-„88
rise can mostly be traced to a rise in the interest margin from 0.2 to 1.2 percent and
likewise, the change in interest margin is an important reason for volatility in other years.
Obviously, one can think of smoothing procedures as in Fixler and Reinsdorf (2006), but
we have chosen here not to do that but instead, focus on average growth over a period of

Appendix Table A.1, Call report items used in estimating fee-based output at current and constant prices
Description                                          Period             Item                             Notes
Fees & commissions
Fiduciary activities                               1987-2004         RIAD4070
Service charges on deposit accounts                1987-2004         RIAD4080
Investment banking, advisory, brokerage, and
underwriting fees and commissions                  2001-2004         RIADB490           Before 2001 based on Census
Net servicing fees                                 2001-2004         RIADB492           Before 2001 based on servicing assets
Net securitization income                          2001-2004         RIADB493           Before 2001 based on servicing assets
Insurance commission fees and income               2001-2004 RIADC386+RIADC387 Before 2001 based on Census
Other noninterest income                           2001-2004         RIADB497           Before 2001 calculated as residual

Holding gains/losses
Trading revenue                                         1996-2004        RIADA220
Noninterest income on other gains (losses) and fees
from foreign exchange transactions                      1987-1995        RIAD4075        Part of RIADA220
Noninterest income on other foreign transaction gains
(losses)                                                1987-1995        RIAD4076        Part of RIADA220
Noninterest income on other gains (losses) and fees
from trading assets and liabilities                     1987-1995        RIAD4077        Part of RIADA220
Net gains (losses) on sales of:
Loans and leases                                        1991-2004        RIAD5416        1)
Other real estate owned                                 1991-2004        RIAD5415        1)
Other assets (excluding securities)                     2001-2004        RIADB496
Premises and fixed assets                               1991-2000        RIAD5417        Almost consistent with RIADB496 1)

Securitized loans
Outstanding principal balance of 1-4 family residential mortgage loans serviced for others:
Serviced with recourse or other servicer-provided
credit enhancements                                    2001-2004           RCFDB804
Serviced with no recourse or other servicer-provided
credit enhancements                                    2001-2004           RCFDB805
Serviced under a GNMA contract                         1992-2000           RCFD5500            Part of RCFDB804+RCFDB805 2)
Serviced under a FHLMC contract with recourse to
servicer                                               1992-2000           RCFD5501            Part of RCFDB804+RCFDB805 2)
Serviced under a FHLMC contract without recourse
to services                                            1992-2000           RCFD5502            Part of RCFDB804+RCFDB805 2)
Serviced under a FNMA regular option contract          1992-2000           RCFD5503            Part of RCFDB804+RCFDB805 2)
Serviced under a FNMA special option contract          1992-2000           RCFD5504            Part of RCFDB804+RCFDB805 2)
Serviced under other servicing contract                1992-2000           RCFD5505            Part of RCFDB804+RCFDB805 2)
Outstanding principal balance of assets sold and securitized by reporting bank:
Home equity lines                                      2001-2004           RCFDB706            3)
Credit card receivables                                2001-2004           RCFDB707            3)
Auto loans                                             2001-2004           RCFDB708            3)
Other consumer loans                                   2001-2004           RCFDB709            3)
Commercial and industrial loans                        2001-2004           RCFDB710            4)
All other loans and leases                             2001-2004           RCFDB711            4)
1) Unclear whether there is a corresponding Call report code before 1991, so extrapolated using trend in total assets
2) Before 1992 extrapolated using trend in overall mortgage-backed securities from Flow of Funds
3) Before 2001 extrapolated using trend in overall asset-backed securities of consumer loans from Flow of Funds
4) Assumed a constant share of overall securitized loans before 2001

Appendix Table A.2, Time series of fee-generating services output at current prices
of U.S.-chartered commercial banks, billions of dollars, 1987-2004
            Fiduciary Service charges Investment   Net         Net       Insurance     Other
            activities  on deposit      banking servicing securitization activities noninterest
                         accounts                  fees      income                   income
   1987         6.8             8.7              0.5          0.5          0.7          0.1          17.7
   1988         7.1             9.4              0.7          0.6          0.8          0.1          19.4
   1989         7.9            10.3              0.8          0.6          0.9          0.1          22.9
   1990         8.5            11.4              1.0          1.1          1.6          0.2          22.9
   1991         9.1            12.8              1.2          1.4          1.9          0.2          23.8
   1992        10.0            14.0              1.5          1.5          2.1          0.3          28.2
   1993        10.9            14.9              1.8          1.5          2.0          0.3          31.9
   1994        11.8            15.3              2.1          2.0          2.8          0.4          33.0
   1995        12.3            16.0              2.5          2.7          3.7          0.5          36.1
   1996        13.7            16.9              3.0          4.0          5.4          0.6          40.1
   1997        16.1            18.5              3.5          5.5          7.5          0.7          41.3
   1998        18.5            19.8              4.5          8.8         12.1          1.0          47.1
   1999        19.7            21.5              5.6         13.3         18.3          1.5          52.6
   2000        21.4            23.8              7.2         13.5         18.5          2.0          54.1
   2001        20.8            26.5              9.1         11.8         16.2          2.9          53.1
   2002        20.4            29.7              9.2         11.4         19.5          3.4          58.9
   2003        20.8            31.7             10.3         14.2         21.8          3.5          58.8
   2004        22.6            31.9              9.7         14.5         22.0          4.2          58.8
Notes: see data appendix for detailed source and method description.

Appendix Table A.3, Extrapolation options for total securitization revenue of U.S.
commercial banks (billions of dollars)
                                                                       1992      1997         2001
Census benchmarks                                                       4.5       9.1         28.0
Trend in servicing assets                                               3.6      12.9         28.0
Trend in securitized loans                                              5.2      13.7         28.0
Trend in servicing assets for servicing fees & trend in
securitized loans for securitization income                            4.7       13.7         28.0
Notes: total securitization revenue includes net servicing fees and net securitization income. The
2001 column shows the actual Call report data, while the 1992 and 1997 columns show different
extrapolation scenarios.

Appendix Table A.4, Output price indices of fee-generating services of U.S.-
chartered commercial banks, 1987-2004 (1987=100)
            Fiduciary Investment Securitization Insurance                Other
            activities banking     activities   activities             activities
   1987       100.0      100.0       100.0        100.0                  100.0
   1988        96.3      99.1        267.0        106.2                  106.0
   1989       112.0      101.4       242.2        116.1                  111.8
   1990       119.8      102.6       211.7        124.4                  119.2
   1991       120.3      103.9       253.4        130.4                  128.6
   1992       131.1      110.7       189.1        138.7                  135.8
   1993       153.0      110.6       220.9        147.0                  143.1
   1994       164.8      103.8       251.6        154.2                  152.2
   1995       175.2      104.6       337.9        164.0                  161.0
   1996       203.1      104.3       301.2        173.7                  170.1
   1997       233.5      101.1       326.2        183.2                  178.5
   1998       259.4      88.9        327.6        186.9                  183.5
   1999       259.8      78.0        302.6        190.5                  189.2
   2000       285.7      68.0        337.2        195.6                  197.9
   2001       295.3      62.6        332.8        200.2                  206.2
   2002       303.5      60.1        279.9        206.8                  211.9
   2003       302.4      62.1        315.4        214.7                  216.1
   2004       320.6      63.0        292.0        224.5                  222.0
Notes: see data appendix for detailed source and method description.

Appendix Table A.5, Implicit loan balance and deposit transaction deflators, 1987-
2004 (1987=100)
             Demand Time & savings Real Estate Other                     Credit    Commercial
             deposits  deposits               Consumer                   Card      & Industrial
   1987        100.0           100.0            100.0        100.0        100.0        100.0
   1988         92.2           116.3            110.6        104.1        103.5        102.3
   1989         87.9           135.2            115.8        110.4        106.7         97.2
   1990         84.4           144.1            110.1        121.7        101.9         98.9
   1991         81.3           157.9            113.0        134.3         98.4        110.1
   1992         83.8           160.1            110.2        146.0         90.9        146.7
   1993         88.3           158.6            102.0        166.4         81.7         98.7
   1994         88.5           170.3            101.8        186.3         78.2        107.5
   1995         89.2           190.6            106.8        208.0         75.5        109.8
   1996         91.7           224.1            114.1        230.4         74.9        116.7
   1997         90.5           265.4            111.4        255.8         73.8        116.8
   1998         92.0           291.0            120.3        249.1         68.0        125.4
   1999         90.2           329.9            127.4        248.7         58.3        140.0
   2000         88.6           374.3            136.9        245.3         55.5        153.8
   2001         95.3           429.9            151.9        245.6         57.0        168.8
   2002         96.1           475.7            145.8        232.9         60.4        161.7
   2003         91.8           508.8            158.1        225.7         65.1        154.6
   2004         91.5           539.6            165.2        225.0         71.1        139.7
Notes: Index calculated based on changes in the balance-to-count ratio. The change in the balance-
to-count ratio is calculated as the growth rate of the year-average outstanding balance of deposits
or loans minus the change in the BLS activity index, which counts the number of deposit
transactions or the number of loans.

           Appendix Figure A.1, extrapolation of securitization income of U.S.
40                commercial banks (1987-2005), billions of dollars








     1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

                     Trend in securitized loans
                     Trend in servicing assets
                     Census benchmarks ('92 & '97) and trend in servicing assets before '92
                     Trend in servicing assets for servicing fees & trend in securitized loans for securitization income