on the convergence of theory and practice in the ... · on the convergence of theory and practice...
TRANSCRIPT
On the convergence of theory and practice in the measurement of
FISIM
Paper prepared by Athol Maritz (Macroeconomics Research Section, ABS)
for the
Economic Measurement Group Workshop
Sydney November 2012
On the convergence of theory and practice in the measurement of
FISIM1
Abstract
The poor performance of FISIM measures during the GFC has renewed debate about the
conceptual basis of FISIM. What exactly do our FISIM measures measure, and do these
square with any reasonable definition of FISIM? What is a reasonable definition of FISIM?
Should it include compensation for risk? Do the current measures include a proper
compensation for risk, or not? What is a proper compensation for risk, and how would we
measure it? Faced with the ongoing conceptual debate, the difficulty in agreeing a
theoretically correct definition of FISIM, and then in developing a good measure of the
concept, the ISWGNA FISIM Taskforce has asked that countries undertake an empirical
study of the performance of alternative FISIM measures. The Taskforce has provided a
number of criteria on which to judge alternative measures.. These criteria naturally reflect
some underlying view of what FISIM is, and how it should behave through periods of market
volatility. How does this square with the conceptual debate? This paper presents the results of
an Australian study of the performance of different measures of FISIM through the GFC, and
draws some tentative conclusions that bear on the conceptual debate, including the potential
value of distinguishing between expected FISIM and realised FISIM and the implications this
might have for the broader issue of how to measure income in the national accounts.
1. Introduction
The FISIM literature centres on five questions of theory:
What is the credit spread, and should it be included in the definition of FISIM
What default risk is taken by the bank, what is the compensation for this default risk,
and should this compensation be included in the definition of bank FISIM
What maturity mismatch risk is taken by the bank, what is the compensation for this
maturity mismatch risk, and should this compensation be included in the definition of
bank FISIM
In finance, risk is compensated by an expected return premium. Actual returns almost
certainly will differ from expected returns. Should the difference between actual and
expected returns be included in the definition of bank FISIM
1 The author works in the Macroeconomics Research Section of the ABS. The views expressed in this paper are
those of the author and do not necessarily reflect the views of the ABS.
How best to decompose nominal FISIM into price and volume components. Closely
related is the question of how to produce a price index for FISIM for inclusion in the
CPI.
The answers to these questions underpin one of the key challenges facing national
statisticians – how to measure the most important component of output and production for
one of the most important industries in a modern economy. The fact that the answers have
proven elusive is cold comfort to practising statisticians who cannot avoid or postpone
measuring the contribution of the financial sector to the national accounts.
This paper presents a view on these questions of theory and considers the consequent
measurement issues. The paper also reports on an empirical study conducted by the ABS into
the performance of alternative FISIM measures (reference rates) through the GFC.
2. Theoretical issues
2.1 Treatment of the credit spread
Much of the early FISIM literature, and some of the later writings, fail to distinguish clearly
between the credit spread inherent in the interest rates charged by banks and the default risk
experienced by banks. A credit spread exists because of the risk that borrowers may default.
But this does not necessarily mean the bank takes default risk. To illustrate the point,
consider the following example:
A bank makes 10 one-period loans of $100 each. The bank knows one loan will default, but
not which one. Assume that the riskless interest rate is 4.5% and that the bank needs a return
of 5% to cover costs. Then the bank will charge each borrower r%, where 900(1+r/100) =
1000(1+5/100). This implies r% = 16.67%.
Then the credit spread is 16.67% – 5% = 11.67%. Yet in this example, the bank takes no
default risk.
To the extent that the theoretical debate has been resolved at all, it is to recognise that the
dollar value of the credit spread – which might be 11.67% times $1000, or times $900
perhaps, or something in between – should not be included in FISIM.
The main issue is how to ensure the credit spread is eliminated from measures of FISIM.
2.2 Default risk
Banks experience default risk when they are uncertain about how many borrowers will in fact
default, and the theory of finance suggests that banks will be compensated by an expected
return premium that is a function of the non-diversifiable component of this default risk.
Exactly what the expected return premium may be is a difficult question and well beyond the
topic of this paper, but to continue the example, let’s assume it is 1%. Then the bank’s
required return is 6% rather than 5% and it will charge each borrower 17.78% rather than
16.67%.
There is more debate about whether the bank’s compensation for default risk should be
included in the definition of bank FISIM. Since taking default risk is part of what banks do,
and get paid for, the obvious position to take on this question is that the expected return
premium for default risk should be included in bank loan FISIM.2
Measuring default risk – as defined - and determining the expected return premium for
default risk, is difficult and unlikely to be done in practice. For a measurement perspective,
the best way forward would be to develop measures of loan FISIM that de facto include any
expected return premium, whatever it may be. This should be feasible because quoted loan
rates and reported interest receipts would include the expected return premium. This is
discussed further below.
2.3 Actual versus expected defaults
In general there will be a difference between actual and expected defaults.3 This begs the
question of whether FISIM should be defined in terms of actual or expected defaults.
Consider the way banks actually do business. A customer asks for a five-year fixed rate loan.
The bank quotes a rate that will be determined by the credit quality (probability of default) of
the borrower, the five-year point on the relevant yield curve, and a required return to cover
other costs and risks the bank takes in the normal course of doing business. The bank quotes a
credit spread to the customer that reflects the probability that the customer will default. For a
portfolio of loans with an expected default rate, the bank includes a credit spread that reflects
the expected rate of default on the portfolio.
This suggests that the natural concept of FISIM is based on what the bank expects to earn at
the time the (portfolio of) loan(s) is made. Call this expected FISIM, and distinguish it from
actual FISIM which is determined by what actually comes to pass.
The analogy between FISIM and insurance has already been drawn in the FISIM literature
(see for example Fixler and Zieschang (2010) and Hood (2010)) to support the argument that
charging a credit spread is like self-insurance against borrower default. The further point is
that the Insurance Service Charge is defined in terms of expected rather than actual claims,
with the difference between expected and actual claims accounted for as a current transfer.
Analogously, FISIM should be defined (and measured) in terms of expected rather than
actual defaults, and the difference between actual and expected defaults should be accounted
for as a capital transfer (see Hood (2010) for a discussion of this issue).
2 Though the contrary view and the ongoing debate on this issue as summarised for example in Zieschang
(2012) is acknowledged. 3 At least in a simple one-period example. More generally, one would expect that actual defaults equal expected
defaults on average over time, if not in every period
Expected FISIM is likely to be a relatively well behaved series over time, and actual FISIM
to be more volatile, particularly through periods of economic difficulty. Assuming expected
FISIM is the relevant concept, the main measurement problem here – as it often is in finance
– is that one observes actual rather than expected returns, and the challenge is to use observed
data to estimate expected defaults.
2.4 Loan FISIM and the role of the reference rate
SNA 2008 defines FISIM and recommends measuring FISIM as the difference between a
loan rate and a reference rate multiplied by a balance, and suggests that the reference rate
should reflect the risk and maturity structure of the loan. This appears somewhat at odds with
the simple example given, where the riskless rate is the natural reference rate.
Consider an even simpler example than that used above – where there is no default risk or
any other risk, and the bank raises money at a riskless rate of 4.5% and lends money at 5% to
cover costs. Then FISIM, equal to the loan rate - here 5% - minus a reference rate - here the
riskless rate of 4.5% - times the loan balance - here 10 times $100 - is earned period-by-
period, and the bank ultimately has principal of $1000 repaid.
Now assume one loan will default, and that the bank charges each borrower a rate of 16.67%
to compensate. The bank earns a margin equal to the loan rate of 16.67% minus the reference
rate of 4.5% times the loan balance of $1000 until the loan defaults, and then the same
margin on a loan balance of $900 thereafter. The important point here is that in order to
equate FISIM earnings in the two examples one needs to subtract $100 of principal from
interest earnings in the second example. More generally one would need to subtract expected
defaults, and in particular expected defaults per period (the default rate), each period. This is
the argument in Hood (2010).
In this example the reference rate is unambiguously the riskless rate, though the same
measure of FISIM could be achieved by adjusting the reference rate for expected defaults.4
2.5 Deposit FISIM and the role of the reference rate
In the simple examples given above, depositors will demand the riskless interest rate plus any
expected return premium for default risk taken by the bank less a payment for deposit
services. Any expected return premium earned by the bank is “passed through” to the
depositors who ultimately take the risk and get the reward (assuming a bank with no equity).
If the same riskless reference rate is used for calculating deposit FISIM as for calculating
loan FISIM, then the recommended method – reference rate minus deposit rate times
balances – will subtract the expected return premium from deposit FISIM. This would be
4 Essentially to include the credit spread. Where the SNA suggests that the reference rate should reflect risk, this
should be taken to mean the credit spread.
incorrect and the appropriate adjustment would require estimation of the expected return
premium.5
The point here is that the expected return premium should be included in loan FISIM – and
this requires that it not be included in the reference rate used for calculating loan FISIM – but
the expected return premium has nothing to do with deposit FISIM – and therefore should be
included in the reference rate used for calculating deposit FISIM. This implies that the
reference rate for calculating deposit FISIM should be a capital market cost of funds.
2.6 CPI FISIM
Consideration of CPI FISIM highlights a major complication in the theory and measurement
of FISIM - while loans are originated in period 0, the payment of FISIM on these loans takes
place in later periods. This raises the question of what is meant by the price paid by
households for the acquisition of FISIM services in period t for CPI purposes, and what is
meant by the dollars earned by banks for the provision of FISIM services in period t for
national accounts purposes.
One view is that the costs of providing FISIM services on a loan originated in period 0 are all
incurred in period 0, in which case it can be argued that all FISIM earned on the loan should
be allocated to period 0 irrespective of when it is actually paid for. For both national accounts
and CPI purposes FISIM in period t would be calculated only on loans originated in period t,
but FISIM would be measured on a Net Present Value basis.
The alternative view is that the cost of providing FISIM services extends over the life of the
loan, and that FISIM is earned period-by-period over the life of the loan. In this case FISIM
earned in period t by banks, and paid in period t by households would be calculated on all
loans being serviced in period t irrespective of when they were originated.
2.7 Maturity mismatch risk
Banks are exposed to maturity mismatch (or term) risk when they borrow short and lend long.
Banks will be compensated in the form of an expected return premium to the extent that there
is a non-diversifiable component to this risk. The bank will charge a fee to borrowers (and
possibly lenders) to compensate for this risk, and this fee should be considered part of FISIM
for the same reasons that the expected return premium for the non-diversifiable component of
default risk should be included in FISIM.6
5 In reality banks do have equity capital and bank shareholders earn the bulk of the expected return premium.
Depositors only bear a residual risk of default, and the expected return premium demanded by depositors is
likely to be relatively small. The expected return premium could be estimated by the difference between riskless
rates and the bank’s cost of funds in the capital markets.
6 Whether there is a non-diversifiable component, and what the expected return would be if there were, are
difficult questions to answer analytically. Given deep and liquid interest rate derivatives markets it may be that
banks are able to hedge much of their maturity mismatch risk if they want to. However if banks can only partly
The theory here seems relatively uncontroversial.7 The more difficult question is whether the
current and proposed methods for measuring FISIM – more particularly the current and
proposed reference rates – properly include the compensation for maturity mismatch risk.
This question has been front and centre of the debate about the merits of multiple reference
rates.
2.8 Multiple reference rates
Abstract from default risk, and assume that $1000 of short-term deposits funds $1000 of
long-term loans. Loan rates are set by bank management with reference to riskless long-term
interest rates (or perhaps swap rates) and deposit rates with respect to riskless short-term
interest rates. A FISIM margin is added to obtain loan rates, and subtracted to obtain deposit
rates. On this characterisation, it seems clear that expected FISIM on loans and deposits
should be determined by multiple reference rates along a riskless yield curve.
However, if one assumes that the difference between long-term rates and short-term rates is a
measure of the expected return premium for maturity mismatch risk, it seems equally clear
that the use of multiple reference rates eliminates this expected return premium. Put another
way, in this simple example (no defaults) FISIM is the margin between long-term interest
receipts and short-term interest payments. Since the use of multiple reference rates eliminates
much of this margin, it can’t be giving a proper measure of FISIM. By contrast, any single
reference rate would give the whole margin.
In terms of reconciling these views, first note that the difference between long and short rates
should not be interpreted as the compensation for taking maturity mismatch risk. A large
literature on the term structure of interest rates recognises that the shape of the yield curve is
largely determined by inflation expectations. Long rates are higher than short rates because
inflation, and therefore short rates, are expected to increase over time. There are times when
short rates are higher than long rates and this is because inflation, and therefore short rates,
are expected to fall over time.
The important point is that term structures are not static or fixed – they are expected to evolve
over time, and the expected evolution of the term structure is implicit in today’s term
structure. One implication is that over time short rates are expected to have an average close
to the relevant long rate. (This is the basis of an interest rate swap transacted with no upfront
payments or receipts. The average floating rate over the life of the swap is expected to be
approximately equal to the fixed rate). Only the difference between the fixed (longer term)
rate and the average of the floating (shorter term) rates – if a difference exists - can therefore
be interpreted as the compensation for term risk.
hedge this risk using interest rate derivatives, and if the residual risk has a non-diversifiable component, banks
will demand an expected return premium to compensate for this risk.
7 Though note that depositors as claimholders of the bank will also demand an expected return premium to
compensate for that part of the risk that they bear. Depositors bear risk to the extent that the bank may default
and fail to repay depositors
The implication is that on average over time the difference between loan rates and a long
term reference rate would be approximately the same as the difference between loan rates and
a short term reference rate. On average over time FISIM would be approximately the same.
However FISIM is likely to be more variable using short term reference rates as those short
term rates move from lower than the long term rate to higher than the long term rate.
In summary, the use of a short term reference rate for long term loans (a single reference rate)
does not serve to ensure that the current difference between long and short rates is captured as
FISIM – because those short term rates are expected to rise in the future. However, neither
should it because the difference between long and short rates is not a measure of FISIM.
The remaining question is whether the use of a long-term reference rate for long-term loans
and a short-term reference rate for short-term deposits (multiple reference rates) eliminates
the expected return premium for maturity mismatch risk, if indeed an expected return
premium does exist.
If there is a non-diversifiable maturity mismatch risk, and if the expected return premium is
implicit in the existing yield curve (so that for example long-term rates are higher than they
would otherwise be), the bank will capture its expected return premium if it prices off this
yield curve. However using (multiple) long-term and short-term reference rates to measure
FISIM will eliminate the premium. If just short rates are used the premium will show up in
loan FISIM; if just long rates are used the premium will show up in deposit FISIM.8
Together with the practical difficulties involved in implementing a multiple reference rate
approach (discussed in section 4.1 below), this supports a single reference rate methodology.
2.9 Summary of the theoretical position
A summary of the theoretical position is:
The credit spread should not be included in FISIM
The expected return premium for default risk should be included in loan FISIM
The expected return premium for maturity mismatch risk should be included in loan
FISIM
Loan FISIM in period t could be defined in terms of loans originating in period t only,
or all loans being serviced in period t irrespective of when they were originated
Riskless rates are the natural reference rates for loan FISIM and cost of funds are the
natural reference rates for deposit FISIM
Multiple reference rates will (erroneously) subtract the expected return premium for
maturity mismatch risk.
8 And in this simple example a weighted average reference rate would distribute the expected return premium
between loan and deposit FISIM. More generally weighted average reference rates have been advocated as a
way of capturing (some of) the benefits of multiple reference rates without eliminating the expected return
premium for maturity mismatch risk.
3. Measurement of FISIM
3.1 Naïve approaches to the measurement of FISIM
The definition of loan FISIM as the product of the margin between the loan rate and the
reference rate and the loan balance seems uncontroversial. The difficulties are all in the
implementation – what loan rates to use, what reference rates and what balances.
It is reasonably obvious that quoted loan rates – or indicator rates – include the credit spread
and that their use as loan rates will overstate loan FISIM. The obvious alternative would be to
rely on reported interest received rather than indicator rates. Reported interest received in
period t is divided by the reported balance in period t to give the loan rate in period t. FISIM
would then be written as:
FISIM = [interest received/balance – reference rate] * balance … (1)
or
FISIM = interest received – [reference rate * balance] … (2)
The obvious problem with this approach is that the reported interest in period t divided by the
reported balance in period t in equation (1) merely recreates quoted interest rates that include
the credit spread – an expected default rate should be subtracted.9 Furthermore, if the relevant
concept is expected rather than actual FISIM, then FISIM on loans made in period 0 should
be based on expected rather than actual balances over the life of the loan.10
The same key point can be made with reference to equation (2). Interest received should be
adjusted for the principal amount of expected defaults. This is the point made in section 2.4
and elaborated by Hood (2010).
The important bottom line for measurement purposes is that information on default rates is
required for a proper measurement of FISIM.
3.2 The problem of unexpected changes in reference rates over time
FISIM in period t on a portfolio of loans originated in period 0 should be measured with
respect to the period 0 reference rate (in other words, the second term in equation (2) above
should be the period 0 reference rate times the period 0 balances). However the naïve
implementation of equation (2) relies on the period t reference rate. This is a problem if
reference rates change. Refer again to the simple example of section 2.1. If riskless rates
change from 4.5% in period 0 to 4% at in period t, the naïve approach will inflate FISIM on
9 Note that there would not be the same problem for non-performing loans carried at, and reported at, book
values. 10
Though in the steady state actual defaults and actual balances will equal expected defaults and expected
balances on average over time
loans by the difference – 0.5% times balances. In periods of interest rate volatility this effect
will translate into (spurious) volatility in the FISIM measure.11
Two methods have been suggested for addressing this problem. The first is to use reference
rates that are appropriately-weighted averages of past reference rates. This is difficult to
implement in practice.12
The second is to base the FISIM measure on period t reference rates
and the period t market value of loans (Fixler and Zieschang (2010), p22). The formula is the
following modification of equation (2) above:
FISIM = interest received – [period t reference rate * period t market value of balance]
The second term is an approximation to the period 0 reference rate * period 0 balances (ie
book values).
3.3 Measurement of FISIM in the Australian National Accounts
3.3.1 Compilation of the interest matrix
The first step in the calculation of FISIM in the national accounts is the compilation of an
‘interest matrix’ showing the flow of interest between institutional sectors. For most sectors
the total interest paid and received is available directly from source data such as APRA. For
the most part, however, inter-sectoral flows are not available from source data, though there
are some exceptions such bank housing interest paid. Inter-sectoral interest flows are
modelled using RBA indicator rates together with balances available from the Financial
Accounts. The indicator interest flows are then pro-rated to ensure the aggregate interest flow
is consistent with the control totals reported in the APRA data. In other words, the inter-
sectoral interest flows are benchmarked to the aggregate interest flows.
3.3.2 Calculation of FISIM
Given interest flows and balances it is possible to calculate rates of interest on loans and
deposits. These rates are referred to as “effective” rates to distinguish them from contractual
or indicator rates.
The ABS uses the midpoint between the effective rate on loans and deposits as the reference
rate. FISIM on loans is the effective rate on loans minus the reference rate multiplied by loan
balances. FISIM on deposits is the reference rate minus the effective rate on deposits
multiplied by the balance on deposits.
11
The volatility is spurious because it is a mark-to-market effect that should not be reflected in the measure of
bank output and production. As argued above, the proper measure of bank output and production is an expected
FISIM measure. 12
Though note that using a reference rate equal to the midpoint between loan and deposit rates effectively
achieves this end. See the discussion in section 3.3.
Figure 1 plots nominal FISIM over the past 10 years, including the GFC period, for all
sectors and for the household sector.13
(Loan and deposit FISIM is shown separately in
Appendix 1).
Figure 1 Nominal FISIM for all sectors and for households
For both households and all sectors, FISIM rose steadily over the period, but accelerated at
the onset of the subprime crisis in mid-2007 for household FISIM, and from the onset of the
GFC in mid-2008 for total FISIM. In both cases FISIM rose over 50% between March 07 and
March 09. In a midpoint methodology for FISIM, the only factors that influence time series
variability of FISIM are changing margins between loan and deposit rates and changing
balances. The charts in Appendix 2 show that margins rose about 15% from Sep 08 to March
09, with the bulk of the rise in nominal FISIM due to an increase in balances.
3.3.3 Assessment of the ABS methodology
The ABS methodology is a naïve approach in that it fails to subtract expected defaults from
reported interest received and it uses reported balances for period t rather than original
balances from period 0. It therefore fails to remove the credit spread.
However the use of midpoints as the reference rate largely mitigates the spurious volatility
due to unexpected changes in interest rates (see section 3.2). This is because midpoints are
(by definition) based on reported loan and deposit rates that in themselves are weighted
averages of past loan and deposit rates. The lower volatility of FISIM based on midpoints is
evident in empirical analyses such as that reported below comparing the performance of
different reference rates through the GFC. These consistently show that FISIM based on
midpoints is less volatile than FISIM based on exogenous reference rates.
13
The household sector here includes owner-occupied housing
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
Nominal FISIM ($m)
All Sectors Households
3.3.4 Price and volume decomposition
The ABS calculates a Laspeyres Chain Volume Measure (CVM) for FISIM in two steps.
Nominal balances are first deflated by the IPD for Domestic Final Demand to remove the
impact of prices on balances. The Laspeyres CVM for FISIM then employs margins as prices
and deflated balances as quantities.
Figure 2 plots nominal and volume FISIM over the past 10 years, including the GFC period.14
Figure 2 Nominal and volume measures of FISIM for all sectors
3.4 Measurement of FISIM in the Australian CPI
The ABS also compiles a price index for FISIM that was included in the CPI until the recent
16th
series review of the CPI, when it was removed from the headline measure on the basis of
concerns about its impact on the CPI through the GFC.15
The conceptual approach to compiling the price index for FISIM in the CPI is similar to that
used in the national accounts, and the CPI price index for FISIM therefore suffers the same
14 The quarter-by-quarter volatility in volume FISIM evident in this chart is due to the fact that nominal FISIM
is first calculated on an annual basis and a quarterly current price series is obtained by diving annual FISIM by
4. However the CVM is calculated on a quarterly basis. This introduces a (spurious) volatility into the quarterly
volume measures.
15 However note also that the price index for FISIM continues to be included in an analytical series. The ABS
plans to include FISIM in the CPI once measurement concerns have been addressed.
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
Total FISIM for All Sectors
Nominal ($m) Volume
limitations as the national accounts measure of FISIM. The main difference is in the data
sources employed. While the national accounts compiles interest flows and balances in
aggregate for each institutional sector, the CPI collects interest flows and current balances by
product from a sample of banks.16
Figure 3 plots the CPI price index for FISIM. The chart also includes the IPDs for FISIM
from the national accounts.
Figure 3 The CPI price index and national accounts IPDs for household FISIM
The CPI measure rose and fell sharply through the GFC. This is consistent with the rise and
fall in the margin between loan rates and deposit rates reported in section 3.3.2. Interestingly,
the IPDs for FISIM from the national accounts show markedly different behaviour – they rise
but do not fall. This may be due to scope differences – the CPI measure only covers
household FISIM, though it does include FISIM paid by owner-occupied households.
Nevertheless, the difference raises some concern about the current methodology for
generating volume measures of FISIM in the national accounts and illustrates the importance
of better integrating the measures of FISIM used in the CPI and the national accounts.
4. Empirical tests of different reference rates
In response to the ongoing international debate about the proper measurement of FISIM, the
Inter-secretariat Working Group on National Accounts (ISWGNA) FISIM Task Force
16
There are also minor differences in the detail – for example, the CPI calculates a separate reference rate for
each bank.
80
90
100
110
120
130
140
Deposit and loan facilities (CPI) IPD
requested that member countries conduct an empirical analysis of the performance of
different approaches to the measurement of FISIM through the GFC.
4.1 Australian results
The following analysis compares FISIM based on the (endogenous) midpoint methodology
and a series of exogenous reference rates: the short term interbank rate, the 5-year
government bond rate, a weighted-average rate and maturity-matched (multiple) reference
rates. The analysis focuses on bank FISIM for the household sector.17
The measurement of FISIM using a single reference rate is relatively straightforward. The use
of multiple reference rates is complicated by the limited availability of interest flows and
balances split by maturity. The Australian data do provide balances by product for the
household sector, which enables an approximate split between long-term and short-term
balances. For example, transaction deposit accounts are readily classified as short term and it
is possible to make reasonable assumptions about the split of long-term (>1 year) and short-
term (<1 year) term deposits.
Interest flow data is more problematic because there is no product detail for interest flows
that would enable a direct measurement of interest flows on long-term products versus short
term-products. The ABS has used an approximate method based on indictor rates reported on
the RBA bulletin. Given the long-term and short-term balances, total interest flows are
calculated based on the indicator rates. These flows are then benchmarked to control totals
for total interest reported.18
Figure 4 shows the different reference rates, together with the effective loan and deposit rates
for households.
17 The main results are presented for the household sector a) to facilitate comparison with the CPI analysis in
section 4.3, b) because data on maturity is much better for the household sector, and c) because the
results/conclusions are similar for the household sector alone and all sectors.
18 Essentially, the calculated long-term and short-term effective rates are proportional to long-term
and short-term indicator rates.
Figure 4 Alternative reference rates and effective loan and deposit rates for households
The time profile of alternative reference rates is similar until mid-2007. All rates rise in the
period leading up to the GFC and then fall sharply. However, the fall in exogenous rates is
more exaggerated than the fall in loan and deposit rates, and the midpoint.
Figure 5 shows FISIM calculated using the different reference rates. (Appendix 3 shows the
results for loan FISIM and deposit FISIM separately).
Figure 5 Household FISIM for different reference rates
0
2
4
6
8
10
12
Alternative reference rates and effective rates - households (%)
RR - 5 year Gov Bond Deposit
Loans RR - interbank lending rate
midpoint Weighted avg RR
0
2000
4000
6000
8000
10000
12000
14000
Total FISIM - Households ($m)
Midpoint interbank lending rate 5 yr Gov Bond rate
Maturity matched Avg RR
FISIM calculated using exogenous reference rates is notably more volatile during the GFC
than FISIM calculated using the midpoint reference rate. The additional volatility is due to
the varying ‘distance’ of the exogenous rates from the effective loan and deposit rates (see
figure 4).19
As discussed in section 3.2, this distance varies because exogenous rates are
current (period t) rates while the effective rates on loans and deposits are weighted averages
of past (period 0) rates.
FISIM calculated using exogenous reference rates is also notably higher during the GFC than
FISIM calculated using the midpoint (for the reasons noted in footnote 14). Given that theory
points towards exogenous reference rates this suggests that the midpoint methodology has
understated FISIM during and after the GFC.
Another feature of the Australian results is that all reference rates – including multiple
reference rates – give approximately the same measures of FISIM in the period prior to the
GFC. The reasons are evident from figure 4 – there was very little difference between long
and short rates for much of the period considered.
For households, total FISIM under the maturity matched approach is almost indistinguishable
from FISIM using the interbank rate – even through the GFC – because 80% of loans and
70% of deposits are short-term.
4.2 European results
The Eurostat report “Results on the FISIM tests on maturity and default risk” presents the
results of an empirical study by European countries of different approaches to measuring
FISIM. The study focused on different approaches to the treatment of maturity and default
risk in the calculation of FISIM.
For maturities, countries compared the time series of "implied" loan and deposit rates with
the time series of two possible reference rates: an interbank rate and a weighted average of
long-term and short-term rates, with the weights proportional to balances. Overall, the
weighted average reference rate performed slightly better with respect to reducing the
volatility and incidence of negative measured FISIM. However the results differed from
country to country and were largely inclusive. A limited number of countries were also able
to test a multiple reference rate approach – for the rest suitable data were not available – but
again the results were largely inconclusive.
19 Note that if loan and deposit balances were the same, total FISIM would be the same for all reference rates,
and the only impact of different reference rates would be the split between deposit and loan FISIM. The
reference rate affects total FISIM only if loan and deposit balances are different, which is generally the case. In
the normal case where loan balances are greater than deposit balances, a reference rate closer to the effective
rates on deposits and further from the effective rate on loans will result in higher FISIM.
Countries also attempted to correct for "expected losses on loans going into default" by
accessing data on write-offs and provisions for bad and doubtful debts. However Eurostat
reports that the results were “inconclusive” due to the limited availability of suitable data.
4.3 Comparison of national accounts results with CPI results
A similar study of alternative reference rates has been conducted by the ABS for CPI FISIM
(Barosevic, Conn and Cullen (2010)). For completeness, the key chart from this report is
reproduced in figure 6 below. In this chart, the cost of funds reference rate is a weighted
average of retail deposit rates, short-term wholesale (money-market) funding rates and long-
term wholesale (AA-rated corporate bond market) funding rates.
Figure 6 The CPI price index for FISIM under three different reference rate models
Consistent with the previous analysis, the use of a midpoint reference rate produces a
substantially less volatile price index for FISIM than the use of exogenous rates.20
20
There is a slight difference in the CPI FISIM index curves in figure 6 and figure 3. This is due changes in the
data used to calculate the index.
References
Fixler, D.J. and Zieschang, K.D. (2010), Deconstructing FISIM: Should Financial Risk
Affect GDP? Paper prepared for the 31st General Conference of the International Association
for Research in Income and Wealth
Hood, K.K. (2010), Computing Nominal Bank Services: Accounting for Default, Paper
prepared for the 31st General Conference of the International Association for Research in
Income and Wealth
Zieschang, K.D.(2012), FISIM Accounting, IMF Working Paper
Appendix 1 Loan and Deposit FISIM
Figure 1.1 Nominal Loan FISIM for households and all sectors
Figure 1.2 Nominal Deposit FISIM for households and all sectors
0
2000
4000
6000
8000
10000
12000
Loan FISIM ($m)
All Sectors Households
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
Deposit FISIM ($m)
All Sectors Households
Appendix 2 FISIM margins and balances
Figure 2.1 Effective loan and deposit rates and the margin between the two - households
Figure 2.2 Loan and deposit balances - households
0
2
4
6
8
10
12
Inte
rest
rat
e (
% p
.a.)
Effective loan and deposit rates and the margin between the two - households
r(D) r(L) difference
0
200000
400000
600000
800000
1000000
1200000
1400000
Balance for loans and deposits ($m) - Households
Loan Deposit
Figure 2.3 Effective loan and deposit rates and the margin between the two – all sectors
Figure 2.2 Loan and deposit balances – all sectors
0
1
2
3
4
5
6
7
8
9
10
Inte
rest
rat
e (
% p
.a.)
Effective loan and deposit rates and the margin between the two - all sectors
loan deposit Difference
0
200000
400000
600000
800000
1000000
1200000
1400000
1600000
1800000
Balance for loans and deposits ($m) - all sectors
Loan deposit
Appendix 3 Loan and deposit FISIM for households using alternative reference rates
Figure 3.1 Loan FISIM – households – using alternative reference rates
Figure 3.2 Deposit FISIM – households – using alternative reference rates
0
2000
4000
6000
8000
10000
12000
14000
Loan FISIM - Households ($m)
Midpoint interbank lending rate 5 yr Gov Bond rate
Maturity matched Avg RR
-500
0
500
1000
1500
2000
2500
3000
3500
Deposit FISIM - Households ($m)
Midpoint interbank lending rate 5 yr Gov Bond rate
Maturity matched Avg RR