Chapter

2 The Morning After: Explaining the Slowdown in Japanese Growth

Author(s):
Tamim Bayoumi, and Charles Collyns
Published Date:
March 2000
Share
  • ShareShare
Show Summary Details
Author(s)
Tamim Bayoumi

What explains the Japanese economic slump of the 1990s? This question has gained importance since the economy’s plunge into recession in early 1997. Before the latest bout of weakness, many regarded the downturn in activity that followed the bursting of the asset price bubble in 1991 as following a normal cyclical pattern, although somewhat longer than usual due to the size of the asset deflation. In particular, the nascent signs of economic expansion through much of 1996 and early 1997 appeared to confirm that the economy was regaining its balance (albeit assisted by some demand shifting in anticipation of the consumption tax hike in April 1997), and could be expected to recover steadily over the next few years.

Rather than recovering, however, in 1997 the economy entered its first recession since the early 1970s from which activity has still not fully rebounded. Combined with the earlier weakness, this means that by 1999 Japan has been in a slump for almost eight years. Growth averaged only slightly over 1 percent a year over the 1991–99 period, and the output gap is estimated to have moved from plus 4½ percentage points of potential output in late 1990 to minus 7½ percent by late 1998.1 This amounted to the most serious economic slowdown experienced by any major industrial country since the early 1950s. Furthermore, this slump occurred despite significant countercyclical policies, involving a considerable expansion in the fiscal deficit (partly through packages aimed at fiscal expansion) and a reduction in the official discount rate to a record low of ½ of 1 percent in September 1995.

The proximate causes of the initial slowdown in output in the early 1990s are generally agreed. In mid-1989 the Bank of Japan started to raise interest rates so as to cool the asset price inflation that had started in the mid-1980s. The tightening of monetary policy pricked what was later identified as an asset price bubble, and stock and land prices started falling rapidly. Just as the run-up of asset prices in the upswing of the bubble had encouraged domestic spending and driven the economy significantly above potential output, so the collapse of asset prices lowered domestic demand and output, and the economy grew at an annual rate of 1 percent or less through 1994.

As the Japanese slowdown turned from temporary slowdown to slump, however, its causes came under further scrutiny, and a number of competing hypotheses emerged. They fall into four main categories. The first is that the slump reflected inadequate policy responses, particularly as regards fiscal expansion (Posen, 1998). Although the Japanese government unveiled a number of fiscal packages aimed at reviving the economy over the 1990s, the argument goes, most of these packages contained limited amounts of “real water” (that is measures that have a direct impact on activity). The main exception was the September 1995 stimulus package, to which the economy responded vigorously until the recovery was derailed by a switch to fiscal contraction in early 1997.2 The implications of this analysis is that fiscal policy is effective, the downturn reflects the normal cyclical factors, and it is the absence of sufficiently bold fiscal policies that explains the length of the Japanese recession.

An alternative view, which focuses on monetary policy, holds that Japan has been stuck in a liquidity trap (Krugman, 1998).3 Consumption is historically low in Japan, creating a high structural saving rate, which was offset during the golden years by high investment. However, a slowdown in anticipated growth has led to a sufficiently large imbalance between saving and investment that the equilibrium real interest is now negative. The anti-inflationary reputation of the Bank of Japan is sufficiently strong that expectations of future inflation are low. As a result, despite record low nominal short- and long-term interest rates, the monetary authorities are unable to reduce the real interest rate sufficiently far to bring the economy back to full employment. Under this analysis, monetary policy is the most effective instrument for countercyclical policy (while past fiscal stimulus has had limited impact because of Ricardian effects), but as a result of the anti-inflationary credentials of the Bank of Japan it has lost traction and is hence unable to pull the economy out of its slump.4

A third view holds that the slowdown has reflected the low rate of return to capital owing to overinvestment (Ando, 1998). Japan is in a vicious cycle, in which past overinvestment is reducing the rate of return on capital, which both lowers current investment and spurs saving, as consumers fail to achieve their desired level of asset accumulation. The usual wealth effects that cause cyclical downturns are being elongated by the inefficiency of the corporate sector, exacerbated by a significant corporate debt overhang that further reduces the incentive to invest. In the absence of wealth-creating investment opportunities, the economy will remain depressed.5 This view gives primacy to wealth effects (largely through the stock market, as land prices have divergent effects on property owners and those with no land) in explaining the prolonged slowdown in Japan, while the structural nature of the imbalance between saving and investment explains and the inability of countercyclical policies produces a significant private sector response.

A final view holds that the slump reflects problems with financial intermediation. Banks play a much more important role in financial intermediation in Japan than in Anglo-Saxon financial systems such as the United States or United Kingdom,6 and are the main providers of loans to small and medium-sized enterprises. During the asset price bubble, the banks lent large amounts of money to firms using land as collateral. With the dramatic fall in land prices since the bursting of the asset price bubble, many of these loans have stopped performing. The bubble in stock prices further exacerbated these effects by first boosting and then reducing bank capital.7 Lax accounting rules and a permissive regulatory environment have allowed banks to survive, but with only limited ability to lend to companies because of the competing need to write off bad loans and maintain capital adequacy ratios.8 As a result, the most productive parts of the corporate sector have been starved of new loans, and small and medium-sized companies in particular have thus been unable to play their usual role as leading sectors in the economic recovery.9 Under this view, lending constraints have been the major constraint on the economy, and the inability to obtain finance has limited the effectiveness of monetary policy (which largely operates through banks) and the ability of the private sector to respond to fiscal stimulus.10

These explanations are not mutually exclusive. Indeed, it would be unlikely that a slump of the type experienced in Japan would have a single cause. However, each explanation points to a different set of variables as the major factor explaining the slump.11 They also point to different explanations of the mini-revival in 1996 and early 1997, with the fiscal and liquidity trap explanations pointing to fiscal stimulus and falling real interest rates (and real exchange rates), respectively, as the main cause of this upturn. The other two explanations, on the other hand, imply that the economy continued its stagnation over this period. The logical corollary is that the upturn was largely illusory, with continuing weakness being obscured by a shift in demand in anticipation of the April 1997 consumption tax hike. These “structural” explanations also imply weakness in different components of domestic demand. If wealth effects are a major cause of the slump, one would expect to see significant movements in consumption in addition to investment, while if the problems are mainly due to financial intermediation, it is more likely to be seen in business investment and, to a lesser extent, residential investment given the limited access of Japanese consumers to bank loans.

This chapter examines the reasons for the slowdown in activity in Japan empirically using vector-autoregressions (VARs) involving the main competing explanations: fiscal policy, monetary policy (including the exchange rate), domestic asset prices, and lending to the private sector. A VAR approach was chosen for a number of reasons. It allows the variables underlying the alternative explanations to be incorporated into a single empirical approach. For example, their impacts on output can be compared using the relevant impulse response functions. In addition, estimating a system of equations allows interactions between different variables to be examined, in particular the relationship between domestic asset prices, lending, and output, as well as allowing changes in underlying behavior to be assessed through examination of the residuals from individual equations. Finally, the historical role of each variable can be examined using the decomposition of past movements in output implied by the VAR.

Past Trends

Before discussing more formal analysis of the causes of the slowdown in Japan in the 1990s, it may be useful to look at the underlying data for output and for domestic demand and its components over the period since 1980.12 As can be seen in Figure 2.1, output has gone through a number of cycles over the last two decades, following a relatively stable growth path from 1980 through 1987, expanding rapidly through the next few years to 1991, and then stagnating from then through early 1995. This is followed by a very limited recovery through early 1997, and a renewed collapse in output that was still continuing in late 1998.13 The figure also shows the IMF’s estimate of potential output based on a Cobb-Douglas production function and the resulting path for the gap.14 The path shows the cyclical path of the economy even more clearly, including the cyclical peaks in 1990–91 and 1997, and troughs in 1983, 1995, and the current downturn.

Figure 2.1.Output and Demand Developments

(Trillions of 1990 yen; Logarithmic scale)

Sources: Nikkei Telecom; WEFA; and IMF staff estimates.

The advantage of correcting for potential output is that it provides a path for the cyclical element in output, which is primarily affected by short-term factors such as changes in aggregate demand. Given the prolonged stagnation of output in Japan, however, any estimate of the path of potential output is highly uncertain. The IMF staff estimate takes account of the impact of changes in business investment on underlying growth, and of demographic changes, in particular, the slowing of growth in the workforce over recent years. Because the calculation is based on a production function, it is less affected by the end-point of the data than other, more statistical approaches, such as a Hodrick-Prescott filter. This is particularly important in this exercise, as the data set ends in 1998 with the Japanese economy in the midst of a recession, so that any procedure that attempts to detrend based solely on the path of output will tend to underestimate the size of the output gap.15 Indeed, the concern with the IMF staff estimate of the output gap used here is the opposite, namely that it may take too little account of certain underlying factors that may have lowered the growth in potential output over the 1990s, such as reductions in the rate of return on capital caused by the excesses of the investment boom over the bubble years or the possible slowing of the rate of technological progress because of inefficiencies in the allocation of capital.16 Despite such uncertainties, the path of potential output provides a useful way of eliminating the underlying supply factors affecting the economy and is used as such in the formal analysis. In any case, the VAR is estimated in first differences, which minimizes the impact of errors in estimating potential output.

The behavior of individual components of demand can also provide insight as to the sources of the recent slowdown in demand. Figures 2.2 and 2.3 graph paths of private consumption, business investment, net exports, government consumption, government investment, and residential investment, measured as a ratio to output.17 If the downturn in output during the 1990s largely tracks consumption, then one would assume that it reflected wealth effects of some form, while weakness in business investment would point more toward financial intermediation. To aid comparison, movements in the three major components of demand (private consumption, business investment, and net exports) are measured on the same scale, as are movements in the three more minor components of demand (government consumption, government investment, and residential investment).

Figure 2.2.Major Demand Components

Sources: Nikkei Telecom; WEFA; and IMF staff estimates.

Figure 2.3.Minor Demand Components

Sources: Nikkei Telecom; WEFA; and IMF staff estimates.

Business investment is clearly the most cyclical element of demand since 1980, increasing markedly as a percentage of GDP over the bubble years compared with the period before or since. The underlying trend in investment is also significantly affected by movements in relative prices, with nominal spending staying fairly constant as a ratio to nominal GDP between the early 1980s and the later 1990s, but the corresponding ratio using real values has increased significantly, reflecting a decline in the relative price of investment goods (a reverse trend of this type holds for net exports).18 By contrast, private consumption has been relatively stable as a ratio to GDP over the last two decades, fluctuating within a relatively narrow range between 57 percent and 61 percent of GDP. The alternative calculations illustrate this stability. Measured as a percentage of observed output, consumption is relatively low over the bubble years, while it is relatively high as a percentage of potential output, because most of the variation is in the denominator rather than the numerator. The source of this stability may well be the limited access of individual Japanese to the stock market (household assets are generally held in bank deposits, while banks are important owners of stocks) and to bank loans, the main factors that fueled consumption booms in other economies with asset bubbles and crashes, such as the United Kingdom in the mid-1980s or the Nordic countries in the late 1980s.19

Like their private sector counterparts, government investment has been significantly more variable than government consumption, although in this case it reflects government policies rather than maximizing behavior. The rapid increase in the ratio of government investment to GDP over most of the 1990s reflects the conscious use of government spending to counter weakness in the private sector, with the spike in spending after the September 1995 fiscal package being particularly notable. As might be expected, residential investment shows a significant increase over the bubble period, encouraged by increases in the relative price of land. Also notable is the spike in residential investment associated with anticipation of the consumption tax hike on April 1, 1997, an increase that finds no parallel at the time of the introduction of the consumption tax in early 1989.

It is also instructive to compare Japan’s current slump with that of the United States in the 1930s. As can be seen in Table 2.1, the period leading up to the decline in asset prices and output were quite similar in both countries in terms of economic growth and interest rates, although bank lending grew significantly faster in Japan. The United States then experienced a very rapid fall in output and prices over 1930–33, followed by an equally vigorous recovery associated with higher share prices, increases in bank loans, and negative short-term interest rates. By contrast, in Japan the slump gathered steam over eight years, with growth and inflation declining over the second half of the period compared with the first, together with most financial indicators continuing to deteriorate. The most striking feature of the Japanese slump is not its severity (it has been relatively mild in many respects) but its length, which seems to have been extended far beyond that of a “normal” business cycle.

Table 2.1.Comparison of the United States in the 1920–1930s and Japan in the 1980–1990s
United StatesJapan
1920–291930–331934–371981–901991–941995–981
Growth of real output4.4-8.68.74.51.40.9
Inflation rate-1.7-8.22.71.71.70.6
Commercial paper rate25.02.60.85.83.70.5
Government bond rate4.13.42.86.64.92.6
Central government balance/output0.1-0.2-0.4-3.1-2.2-4.0 3
Share price12.8-8.715.116.4-9.3-6.2
Bank loans2.6-14.11.010.02.8-0.2
Sources: Japanese data from IMF, U.S. data from the National Bureau of Economic Research (NBER) supplied by Michael Bordo.

1995:Q1–1998:Q2 unless otherwise indicated.

Gensaki rate for Japan.

FY1996 and FY1997. The FY1997 data are IMF estimates.

Sources: Japanese data from IMF, U.S. data from the National Bureau of Economic Research (NBER) supplied by Michael Bordo.

1995:Q1–1998:Q2 unless otherwise indicated.

Gensaki rate for Japan.

FY1996 and FY1997. The FY1997 data are IMF estimates.

Econometric Analysis

This section reports the results from VARs using output, two fiscal variables (the structural general government deficit is divided into direct government spending and taxes net of transfers20), two monetary variables (the real short-term interest rate and the real exchange rate), two domestic asset prices (real stock prices and real land prices21), and financial intermediation. (Data sources are provided in the appendix.) Financial intermediation was measured as lending to the private sector by banks, public institutions, and capital markets. As private bank lending turns out to be the most important component, representing over 70 percent of all lending and dominating quarter-to-quarter changes, this series will be simply referred to as bank lending below.22 Output and real bank lending were divided by potential output to eliminate the trends caused by expanding supply, and logarithms were taken of those variables with no clear unit of measurement (the real exchange rate, real stock prices, real land prices, and real lending). In addition to a constant term, the VARs also included two dummy variables aimed at capturing the short-term shifting of demand seen the quarter before and after the introduction of the consumption tax in 1989 and the consumption tax hike in April 1997, with each variable being designed so that the impact sums to zero over time.

The first stage in the analysis involved investigating the statistical properties of the underlying series. The output gap is shown in Figure 2.1, while those of the other explanatory variables are shown in Figure 2.4. Even though most of the series are adjusted by potential output, many still appear nonstationary, with no tendency to revert to an underlying mean value or trend. This even appears to be true of the output gap, despite the fact that output should at some point revert to its level of potential. As discussed earlier, this presumably reflects the depth of the current recession, which makes it appear that deviations from trend can be permanent.

Figure 2.4.Underlying Variables

Sources: Economic Planning Agency (EPA) and WEFA. See text for more details.

Formal analysis confirms these visual impressions. Table 2.2 shows the results from running Dickey-Fuller tests on the various components of the VARs. Almost all of the variables, including the output gap, fail to accept stationarity (without the inclusion of a time trend). Even the two exceptions, bank lending and taxes net of transfers, fail the test when a time trend is included, spectacularly so in the case of real lending. When the variables are first differenced, however, the opposite result holds true, with almost all of the variables accepting stationarity. The exception is real land prices, where the test cannot reject nonstationarity. After some experimentation using both first and second differences, the first difference of land prices was used in the VARs, as this did not appear to cause problems in the estimation.

Table 2.2.Dickey-Fuller Test Results
LevelFirst DifferenceSecond Difference
No trendTrend
Output0.800.950.000.00
Direct government spending0.250.760.000.00
Taxes net of transfers0.030.110.000.00
Real interest rate0.550.070.000.00
Heal exchange rate0.760.840.000.00
Real land prices0.621.000.430.00
Real stock prices0.510.960.000.00
Real bank lending0.051.000.000.00
Memorandum items:
Private consumption0.380.630.000.00
Business investment0.790.980.000.00
Residential investment0.480.820.000.00
Source: Author’s calculations.
Source: Author’s calculations.

It remains possible that there are cointegrating relationships between the levels of the variables, which would imply estimating the equation in levels terms. To investigate this, the model was estimated using the Johansen (1991) procedure, in which the number of cointegrating variables can be tested, although the tests are asymptotic and may not be very robust in the current context given the relatively short sample (under 20 years) and large number of variables. The results of such a test, shown in Table 2.3, indicate a large number of cointegrating relationships. However, none have particularly intuitive properties when normalized with respect to output.23 As an alternative approach, the VAR was estimated in levels terms, which is equivalent to assuming there is one cointegrating relationships for each equation. Again, the estimated cointegrating relationship for output were unsatisfactory and, in addition, the impulse responses from this system exhibited considerable cycling and instability.24 Accordingly, it was decided to focus on VARs using only first differences. Such an approach has the additional advantage that the constant terms in the estimation act as trends, making the estimation less dependent on the assumptions made about the path of potential output.

Table 2.3.Results from the Johansen Procedure
Number of Cointegrating RelationshipsTrace Test90% Critical Valueλ Max90% Critical Value
1237.8150.061.532.3
2176.3117.755.638.4
3120.789.439.624.6
481.164.729.920.9
551.243.823.017.2
628.326.715.213.4
Notes: The test assumes that all variables have integration one and that there are two lags in the underlying VAR.
Notes: The test assumes that all variables have integration one and that there are two lags in the underlying VAR.

Accordingly, a VAR involving the first difference of the output gap, the other explanatory variables, a constant term, and dummy variables for the consumption tax changes of 1989 and 1997 was estimated from the first quarter of 1981 to the first quarter of 1998.25 Two lags were used in the estimation as this was the lag length indicated by the Akaiki Information Criterion.26 A Choleski decomposition was used to orthogonalize the underlying errors using the ordering: direct government spending, taxes net of transfers, the output gap, the real exchange rate, real stock prices, real land prices, and real bank lending. The ordering determines the level of exogeneity of the variables, with changes in government spending being assumed independent of all other explanatory variables, while current changes in bank lending are assumed to be affected by changes in all of the other explanatory variables. The ordering was chosen on the basis of the speed with which the variables respond to current events, with fiscal variables assumed to be the least responsive, followed by output, then monetary policy, asset prices, and bank lending.

The estimated impulse responses for output, shown in Figure 2.5, are generally intuitive.27 The top left panel of the figure, for example, reports the impulse response of the level of output to a one standard deviation shock in direct government spending, together with the level response of direct government spending to its own shock (all of the variables are measured in such a manner that a change of 0.01 represents a 1 percent changes in the relevant variable28). An increase in direct government spending provides the expected temporary boost to the economy while an increase in taxes lowers activity. The dynamic multiplier for direct government spending, calculated using the ratio between the response of output and the response of government investment, indicates that in the short term a ¥100 increase in government spending raises output by about ¥65.29 This stimulus wears off quite rapidly, and after about a year output is estimated to be roughly unchanged. This is consistent with those who have argued that higher government spending has been relatively ineffective over time because uncertainty about the timing and “real water” content of stimulus packages, and the choice of projects with low social rates of return. The implied multiplier from a tax increase, which peaks at -0.2 (in absolute value) after two quarters, is again quite small. This is consistent with the view that Japanese consumers are relatively Ricardian, although the focus on temporary tax cuts in the stimulus packages of the 1990s, rather than permanent changes in the tax system, is also likely to have lowered the multiplier.30 In short, while fiscal policy is effective in stimulating output, the estimated impact is muted.

Figure 2.5.Impulse Response Functions

Source: IMF staff estimates.

An increase in the real interest rate of 1 percentage point lowers output by about 0.6 percent. This is consistent with, although at the lower end of, the wide range of estimates from large models (see Krugman, 1998).31 An increase in the real exchange rate also lowers output in the short term, although the effect is quite small—a 10 percent increase in the real exchange rate lowering output by about 0.2 percent, reflecting the relatively closed nature of the Japanese economy. Output rises in response to an increase in the real price of land and, to a lesser extent, to an increase in the price of stocks. Notably, it also rises quite significantly in response to an increase in bank lending, with a 3 percent increase in such lending leading to a 1 percent rise in output. The absolute size of the various impulse response functions is also illuminating, as they illustrate the impact of a “typical” disturbance in each variable on output. The largest response is associated with land prices, where a typical quarterly disturbance changes output by about 1 percent over time, compared with a value of 0.3 percent for real interest rates.

Figure 2.6 reports the standard errors around the responses of output (calculated using Monte Carlo methods with 500 replications), which indicate that the short-term responses are reasonably well identified. Over longer periods, however, the degree of precision deteriorates, because the underlying impulse responses are cumulated over time, compounding uncertainty.

Figure 2.6.Impulse Response Functions: Output and Standard Errors

Source: IMF staff estimates.

Tests indicate that the results are relatively invariant to alternative orderings of most of the variables. This, however, is not the case for the relative position of land prices to stock prices or to bank lending, because of a significant colinearity between the residuals. If land prices are placed after the other two variables in the ordering, the estimated long-term impact on output becomes similar across all three variables. The ordering chosen was felt to be the most “reasonable,” in that land prices are the least likely variable to be immediately affected by other developments. The nexus of domestic asset prices and bank lending is discussed further below.

To this point, the analysis has focused on the output responses implied by the system. It is also of interest to examine the most important interrelationships between the individual equations comprising the VAR, as these provide information as to the transmission mechanisms at work. These interrelationships can be analyzed through F-tests of the significance of each variable in each equation (Granger causality tests). The results from this exercise again are in accord with intuition. Output is most affected by past changes in real interest rates (note that the main impact of direct government spending is contemporaneous, as government consumption and investment feed through directly into GDP), and least affected by own shocks and real stock prices. Fiscal policy and the real interest rate are relatively independent of the other variables in the model, indicating that government policy decisions are made relatively autonomously, while the real exchange rate is also largely independent of the rest of the model.

By contrast, there are important interactions between stock prices, land prices, and bank lending. As can be seen from the impulse responses in Figure 2.7, positive disturbances in any one of these variables produce increases in all of them. This mutually reinforcing interaction helps explain the asset bubble of the late 1980s. It reflects, at least in part, the importance of domestic asset prices in the behavior of banks, with land being used as the most usual form of collateral, and shareholdings being an important source of bank capital.32 In the 1990s, this process apparently went into reverse, hurting the economy through a reinforcing erosion of bank collateral, capital, and loans (called by some of the more melodramatic commentators the Japanese “death spiral”).

Figure 2.7.Impulse Response of Financial Variables

Source: IMF staff estimates.

The importance of these interactions can be examined by rerunning the VAR with one of the variables exogenized. This is done by excluding the chosen variable from the VAR, but including its first two lags as exogenous variables. The estimated equations for the remaining variables are identical to the main case, but any interactions involving the exogenized variable are no longer identified. When bank lending is exogenized in this manner, the impulse response of land price on output is lowered by almost 90 percent while the impulse response of stock prices falls by two-thirds, implying that the vast majority of the estimated impact of asset prices on output comes through financial intermediation.33 Exogenizing land prices and stock prices in a similar manner also produces significant, if somewhat less spectacular, reductions in the impulse responses of the remaining financial variables with respect to output. In short, there appears to be a close and highly interwoven interrelationship between domestic asset prices and bank lending, an interrelationship that helps to explain the size and longevity of the estimated effects of each of these variables on output.

The cumulated residuals from each equation, shown in Figure 2.8, help to illustrate the direction of the underlying shocks (assuming the shocks are random, they should cumulate to random walks, which have apparent trends over time). In addition to illustrating policy changes (such as the spike in government spending after the September 1995 stimulus package was announced and the tightening of monetary policy in late 1989), the results also illustrate the rise and fall in domestic asset prices over the bubble and subsequent crash, the increase in bank lending in the early 1980s (a time of significant deregulation) and more recent weakness, and the large positive shocks to output in 1996, prior to the consumption tax hike.

Figure 2.8.Cumulative Residuals

Source: IMF staff estimates.

The decomposition of past movements in output implied by the model is shown in Figure 2.9. Past changes in the output gap are divided into those parts explained by innovations in fiscal policy (the sum of direct government spending and taxes net of transfers), monetary policy (the sum of real interest rates and the real exchange rate), asset prices (the sum of land prices and share prices), bank lending, and exogenous disturbances (the sum of independent shocks to output, the dummy variables, and any effects due to unidentified disturbances prior to the estimation period). Note that, as unidentified disturbances to output are included in the “exogenous” term, it is possible to conclude that past movements in output are not very well explained by any of the explanatory variables in the model—the underlying hypotheses can all be refuted.

Figure 2.9.Decomposition of Output

Source: IMF staff estimates.

The decomposition indicates that the most important factor explaining past movements in output is innovations in asset prices, accounting for most of the hump in the output gap over the bubble period and subsequent weakness. Changes in bank lending help to explain the rise in output in the early- to mid-1980s and more recent weakness in activity, indicating that shocks to bank lending can also generate significant movements in output. Monetary policy was supportive though the bubble period, restrictive through much of the 1990s, and more recently again providing a significant boost to the economy. Fiscal policy provided a significant boost to the economy in 1995 and early 1996, but this support was rapidly with-drawn in the later part of 1996. While exogenous factors play an important role in explaining quarter-to-quarter variation in output, they only matter for overall movements in output over the more recent period.

Figure 2.10 decomposes the aggregate fiscal, monetary, asset price, and exogenous effects into their constituent parts (in the case of fiscal policy, for example, the effects of direct government spending and taxes net of transfers are distinguished). They indicate that the fiscal expansion of 1996 was largely fueled by direct government spending,34 mirroring the sharp increase and subsequent fall in government investment, that changes in real interest rates have been the most important monetary policy channel, and that changes in land prices have been generally more important than stock prices in explaining movements in output.

Figure 2.10.Decomposition of Output: Detailed Results

Source: IMF staff estimates.

Possibly the most striking result, particularly when compared with the bubble period, is the significant role played by own shocks in increasing output in late 1996 and early 1997 and reducing it subsequently. The most intuitive interpretation of this is that it reflects longer-term demand shifting from the consumption tax hike rather than that captured by the existing dummy variable, particularly for residential investment (Figure 2.3). This hypothesis is supported by the results from adding a third dummy variable to the model, allowing the 1997 tax hike to affect demand from the beginning of 1996. These estimates (not reported for the sake of brevity) indicate there was a substantial boost to output from early 1996 to early 1997 because of the consumption tax hike, followed by a fall in 1997 and early 1998 (this change in the specification had little impact on other responses).

The historical decomposition also has implications for the differing explanations for the mini-revival of output in 1996. The results shown in Figure 2.9 indicate that fiscal policy and monetary policy both contributed, each providing a boost of about 1 percent to output. However, the underlying situation appears to have started worsening in late 1996 (in part because of a sharp fall in public investment), an underlying weakness that was obscured by demand shifting in anticipation of the consumption tax hike.

The importance of bank lending as a conduit for asset price effects is illustrated in Figure 2.11, which graphs the estimated impact of land prices and stock prices on output once bank lending has been exogenized as described earlier. In this experiment, asset prices produce very limited movements in output, indicating that the “pure” effects of changes in wealth are quite limited.35 A comparison of Figure 2.9 and 2.11 vividly illustrates the central role played by financial intermediation in transmitting asset price shocks to the real economy.

Figure 2.11.Decomposition of Output with Lending Exogenized

Source: IMF staff estimates.

The discussion to this point has focused on a single specification. To examine the robustness of the model, the VAR was reestimated under a number of alternative assumptions. The lag length of the VAR was extended from two lags to three lags, which produced very similar results (with more complex impulse responses). Next, the impact of changing the estimation period was examined, both by truncating the sample at the first quarter of 1996 to avoid the distortions associated with the consumption tax hike in 1997, and by extending the estimation period back to 1973. The VAR was rerun using nominal variables instead of their real equivalents, to examine whether nominal asset price changes produce a more significant impact on the model. Finally, experiments using different proxies for financial intermediation (restricting the variable to cover only bank lending or only lending to the corporate sector) were also conducted. None of these experiments changed the qualitative nature of the results.36

In another type of experiment, additional variables were included in the estimation. First, the old-age dependency ratio was added to the VAR, in order to examine the role of demographic changes in explaining the bubble and subsequent slump. Demographic changes were found to increase output by about ½ of 1 percent over the 1980s and lower it by the same amount over the 1990s. At least some of this effect comes through asset prices, in that increases in the old-age dependency ratio were found to lower domestic stock and land prices, presumably reflecting the reduced demand for such assets from older individuals. The impact on the remainder of the model was minimal. Next, the capital stock was also added to the VAR, to see if a direct measure of overinvestment (the ratio of the capital stock to potential output) helps to explain past changes in output. This variable also had minimal effects either on output or the rest of the model. Finally, real narrow money (Ml) was substituted for the real interest rate to see if a different measure of monetary policy had a significant effect on the results. The money supply provides a good substitute for the real interest rates within the estimation, but has very little impact on the other impulse responses. Exogenizing bank loans in this system still generates significant reductions in the impact on output of shocks to land prices (down by two-thirds) and stock prices (down by one-third), although these effects are somewhat smaller than when real interest rates are used. Hence, even with the money supply included, bank loans still appear to be an important conduit for asset price effects.37

An alternative way of examining the robustness of the results is to consider what happens when variables other than output are used in the VAR. In particular, if the conclusions from this analysis are valid, one would expect that same types of patterns found for output to be apparent in an analysis using the major components of aggregate demand. Accordingly, the VAR was reestimated three times, with output each time being replaced by a different major component of demand (private consumption, business fixed investment, and residential investment).38

The estimated impulse responses for each component of demand are shown in Figure 2.12. The impulse response functions for output from the various shocks appear generally sensible. Increases in government direct spending crowd out private consumption and business investment, but crowd in residential investment, which is what might be expected given the concentration in government investment projections on infrastructure projects of doubtful overall efficiency. Increases in taxes and interest rates lower all of the components of demand, again as might be expected, while the impact of the real exchange rate on domestic demand is small. Finally, increases in stock prices, land prices, and bank loans all raise demand.

Figure 2.12.Impulse Response Components of Demand

Source: IMF staff estimates.

The decomposition of historical movements in private consumption, business investment, and residential investment can be seen in Figure 2.13. The dominant factor explaining movements in business investment and consumption has been asset prices, while residential investment has been largely affected by bank loans, partly offset by expansionary fiscal policy, plausibly reflecting higher government investment (the panels are all produced on the same scale to aid comparisons across different components). Further analysis (not reported) indicates that land prices continue to be at least as important as share prices in explaining the behavior of output, and that bank lending remains an important channel for asset price movements. Hence, the analysis of the components of demand broadly confirms the conclusions of the original analysis.

Figure 2.13.Decomposition of Components of Demand

Source: IMF staff estimates.

Conclusions

This chapter has examined the reasons for the marked slowing of growth in Japan in the 1990s in the context of a VAR analysis that includes the impact of fiscal policy, monetary policy, domestic asset prices, and bank loans. The results are used to attempt to differentiate between a number of alternative explanations of the slump, including the absence of bold and consistent fiscal stimulus, the limited room for expansionary monetary policy because of a liquidity trap, asset price deflation operating through the long-term problems caused by overinvestment, inadequate returns on saving and debt overhang, and disruption of financial intermediation.

The results indicate that all of these explanations have some validity. Fiscal policy has generated limited effects on output except in the wake of the September 1995 stimulus package, whose beneficial effects were rapidly reversed by an abrupt shift to fiscal contraction.

Expansionary monetary policy is also found to have been effective in stimulating demand in late 1997 and 1998, but presumably reached close to its practical limit as short-term interest rates were cut to very low levels. Domestic asset price changes were an important factor behind the rise in the output gap over the bubble period and the subsequent decline. However, the important role assigned to land prices appears inconsistent with explanations that emphasize pure wealth effects as an explanation of the slump (changes in land prices have different effects on individuals depending on whether they own land or not), or with explanations that emphasize structural problems caused by declining rates of return on reproducible capital.

What the analysis reveals is the central role played by financial intermediation in magnifying the impact of asset prices on the economy. Increases in bank lending, operating both directly and through a self-reinforcing cycle with increases in land prices (the main source of collateral) and stock prices (an important component of bank capital), help explain much of the expansion in the output gap in the mid- to late-1980s. The reverse process operated with equal force over the contraction, as undercapitalized banks responded to falling asset prices and other balance sheet pressures by restraining lending to maintain capital adequacy standards.

The importance of banks both in overall lending and, in particular, in providing capital to smaller companies, which have failed in their usual role of leading the economy out of recession, provides an obvious mechanism through which domestic asset prices and bank lending could have disrupted activity. The central role played by financial intermediation in the slump also generates a compelling reason for the limited effectiveness of standard macroeconomic policies. If the corporate sector is limited in its ability to obtain funds, then this will blunt the impact of monetary policy (as such policy operates largely through the banking system) and of fiscal policy (as companies and individuals will be constrained in their ability to respond to government stimulus). Finally, it provides a ready explanation for the recession in 1997–98. Already undercapitalized banks responded to the prospect of tighter banking regulations in early 1998 (when “prompt corrective action” was introduced) by further cutting back on lending, exacerbating the weakness already generated by fiscal contraction and the Asia crisis, and sending the economy rapidly into the doldrums.

At the same time, the limitations of this exercise should be borne in mind. VAR analysis is a powerful tool, but it assumes that the underlying responses are linear and have not changed over time. Both assumptions could be questioned in the context of the type of slump currently experienced in Japan. Individuals could react differently to events depending on the state of the macroeconomy, with behavior at the tip of a cyclical upturn being rather different from that at the bottom of a downturn. Similarly, the impact of financial sector deregulation since 1980 may have altered the relationship between the corporate sector and the banking system.

More analysis, looking more deeply at the mechanisms through which the banking system might affect output, would be needed to support the results from this paper (see Chapter 7 for some subsequent work on this topic). However, the fact that these results appear robust across a number of different specifications provides evidence that banking system problems are indeed at the heart of the current weakness in activity.

Appendix. Data Sources

The sources for the variables were as follows:

Output and components of demand: The National Income Accounts.

Direct government spending: The sum of real quarterly government consumption and public investment.

Taxes net of transfers: Nominal seasonally unadjusted quarterly general government deficit (defined from its components) less unadjusted direct government spending. As the series was not seasonally adjusted and tax policy normally occurs on an annual basis, the series used in the regressions was the four quarter moving average, first differences by subtracting the same value from the year before. Projected after 1997:Q1 due to lack of data.

The real interest rate: The gensaki rate less the inflation rate of the GDP deflator (adjusted for indirect tax changes) over the previous 4 quarters.

The real exchange rate: The IMF’s multilateral real exchange rate calculated using data on unit labor costs across developing countries.

Real stock prices: Monthly averages of the Nikkei 225 index, divided by the GDP deflator.

Real land prices: The average value of land in the six major cities, divided by the GDP deflator. Interpolated from semiannual data.

Real loans: The sum of liabilities of the corporate sector and borrowing by the private sector, as measured by the flow of funds accounts, divided by the GDP deflator.

References

    AndoAlbert1998“Demographic Dynamics and the Causes of the Japanese Recession” (unpublished: Philadelphia; University of Pennsylvania).

    BayoumiTamimChristopherTowe and IchiroOishi1998“Fiscal Policy Issues,” in Japan: Selected Issues IMF Staff Country Report 98/113 (October) (Washington: International Monetary Fund).

    BernankeBen S.1983“Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression,”American Economic ReviewVol. 73 (June) pp. 25771.

    BordoMichaelTakatoshiIto and TokuoIwaisako1997“Banking Crises and Monetary Policy: Japan in the 1990s and U.S. in the 1930s” (unpublished: University of Tsukuba).

    BorioClaudio1996“Credit Characteristics and the Monetary Policy Transmission Mechanism in Fourteen Industrial Countries: Facts, Conjectures, and Some Econometric Evidence,” in Monetary Policy in a Converging Europeed. by K.AldersK.KoedijkC.Kool and C.Winder (Dordecht: Kluwer Academic Publishers).

    DreesBerkhard and CeylaPazarbaşioğlu1998The Nordic Banking Crises: Pitfalls in Financial Liberalization? IMF Occasional Paper No. 162 (Washington: International Monetary Fund).

    GordonRobert1988“Back to the Future: European Unemployment Today Viewed from America in 1939,”Brookings Papers on Economic Activity: 1Brookings Institution pp. 271304.

    HicksJohn1937“Mr. Keynes and the Classics: A Suggested Reinterpretation,”EconometricaVol. 6 pp. 14759.

    JohansenSoren1991“Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregression Models,”EconometricaVol. 59 (November) pp. 155180.

    KeynesJohn M.1936The General Theory of Employment Interest and Money (London: Macmillan).

    KrugmanPaul1998“It’s Baaack: Japan’s Slump and the Return of the Liquidity Trap,”Brookings Papers on Economic Activity: 2Brookings Institution pp. 137205.

    KwonEunkyung1998“Monetary Policy, Land Prices, and Collateral Effects on Economic Fluctuations: Evidence From Japan,”Journal of the Japanese and International Economies 12: pp. 175203.

    LincolnEdward1998“Japan’s Financial Problems,”Brookings Papers on Economic Activity: 2Brookings Institution pp. 34785.

    LipworthGabrielle and GuyMeredith1998“A Reexamination of Indicators of Monetary and Financial Conditions,” in Structural Change in Japan: Macroeconomic Impact and Policy Challengesed. by B.AghevliT.Bayoumi and G.Meredith (Washington: International Monetary Fund).

    OgawaKazuo and Shin-ichiKitasaka1998“Bank Lending in Japan: Its Determinants and Macroeconomic Implications” (unpublished; Osaka: Osaka University).

    OgawaK.S.KitasakaT. H.Yamaoka and Y.Iwata1996“Borrowing Constraints and the Role of Land Asset in Japanese Corporate Investment Decision,”Journal of the Japanese and International EconomiesVol. 10 (June) pp. 12249.

    OgawaKazuo and KazuyukiSuzuki1998“Land Value and Corporate Investment: Evidence from Japanese Panel Data,”Journal of the Japanese and International Economies 12: pp. 13249.

    PosenAdam1998Restoring Japan’s Economic Growth (Washington: Institute for International Economics).

    RomerChristina1992“What Ended the Great Depression?”Journal of Economic HistoryVol. 52 pp. 75784.

    WescottRobert1996“Assessing the Risks of a Credit Squeeze Among Small and Medium-Sized Enterprises in Japan,” in Japan: Selected Issues IMF Staff Country Report No. 96/114 (Washington: International Monetary Fund).

Based on IMF estimates at the time of writing. The adjustments discussed in Chapter 5 have not been incorporated into these data. However, for reasons discussed below, these revisions are unlikely to materially affect the results.

See Chapter 6 for a more detailed discussion of fiscal policy over the 1990s.

Traditional monetarists, however, argue that the liquidity trap is an illusion, and that the problem with policy has been the lack of expansion of the monetary base. This hypothesis will also be considered below.

Some have also pointed to demographic effects in this connection, with the aging population depressing investment more than saving.

See Borio (1996) for a cross-country comparison of financial systems.

Bank capital is susceptible to changes in stock prices because Japanese banks typically hold large amounts of stock in industrial companies. Kwon (1998) explores the relationship between monetary policy, land prices, bank lending, and output using a VAR.

See Ogawa and others (1996) and Ogawa and Suzuki (1998) for evidence on how land collateral has affected investment by Japanese firms.

See Ogawa and Kitasaka (1998) for a discussion of the determinants of bank lending and its possible impact on investment, and Wescott (1996) for a discussion of the role of small and medium-sized companies in the 1995–96 upturn in activity.

Monetary policy transmission is discussed further in Chapter 7.

These explanations also correspond to the alternative explanations of the recovery of the United States from the 1930s’ depression. The fiscal explanation, for example, is favored by Gordon (1988), the liquidity trap by Romer (1992), while the role of financial intermediation is discussed in Bernanke (1983). For a comparison of the role of banks and monetary policy in the two periods, see Bordo, Ito, and Iwaisako (1997).

The year 1980 was chosen as a start for the empirical analysis to ensure that there was a significant period before the bubble economy of the mid-1980s, allowing the extended cycle in output since 1987 to be put in context. As discussed later in the text, extending the period back to 1973 (to the golden period of exceptionally vigorous Japanese economic growth, in which the underlying forces shaping the economy were probably somewhat different than they were subsequently) has little impact on the results.

The data for total domestic demand have a similar pattern, although the period before 1987 looks somewhat less buoyant.

See Chapter 5 for a more detailed discussion of the IMF’s estimate of the output gap. This study was completed before the revisions discussed in that chapter were completed.

Krugman (1998) makes a similar point.

Each component of demand is measured in three ways: nominal spending as a percentage of observed nominal GDP; real demand as a percentage of observed real GDP, which adjusts for changes in relative prices over time; and real demand as a percentage of potential output, which takes account of both relative prices and the cycle.

See also Chapter 3 for discussion of the behavior of components of aggregate demand.

For a more detailed discussion of the experience of the Nordic countries, see Drees and Pazarbaşioğlu (1998).

The fiscal variables are adjusted for the cycle using the IMF’s standard approach, which involves correcting taxes and social security contributions for the impact of the output gap using a buoyancy ratio, government spending for the impact of unemployment (a relatively unimportant effect in Japan, with its limited social safety net), and dividing the relevant totals by potential output.

The role of real land prices in wealth is tricky, as changes in prices affect homeowners very differently from prospective buyers. Land prices can have an aggregate impact on the household sector in some circumstances, even if these groups behave differently from each other. In addition, the price of land is an important consideration for bank loans, as it provides the main form of collateral.

As discussed later, the results are not sensitive to alternative measures of lending.

Different assumptions about the number of lags and the deterministic trends produced similar results.

These results probably reflect the fact that including a levels relationship in a VAR that already has two lags (to provide reasonable flexibility in the dynamic responses) implies estimating a high ratio of parameters to observation (27 parameters from 67 observations). This inevitably lowers the precision of the estimates.

The start date of 1981 reflects the need to accommodate transformations of the underlying data and lags in the VAR.

The Schwartz Bayesian information criterion implied an optimal lag length of three. VARs using a third lag were also estimated, as discussed later.

Note that these responses refer to the levels of output, and so on. As the model was estimated in first differences, underlying disturbances can result in permanent changes in the underlying variables.

This is achieved through measuring the variable in logarithms, as a ratio to potential output or, in the case of the real interest rate, by dividing the percentage value by 100.

These estimates are significantly smaller than the multipliers produced from large models (the IMF, for example, has estimated a multiplier of 1–1.2 in its analysis, see Lipworth and Meredith, 1998).

Bayoumi, Towe, and Oishi (1998) discuss in more detail various reasons how the implementation of fiscal stimulus may have muted its effects over the 1990s.

When the real interest rate was divided into the nominal rate and inflation, it was found that each component was correctly signed and contributed about one-half to the estimated impact.

Kwon (1998), also using a VAR approach, finds that collateral effects increase the impact of monetary policy on the economy, but does not explore the wider set of interactions examined in this paper. Lincoln (1998) provides a detailed discussion of the Japanese system of financial intermediation.

Reversing the ordering of land and stock prices in the VAR leads to a fall of about three-quarters in both impulse responses, when bank lending is exogenized.

By contrast, fiscal contraction in the 1980s largely operated through tax increases.

When the ordering of land and stock prices in the VAR is reversed, the impact of stock prices increases, but still only accounts for a variation in the output gap of around 1 percentage point of GDP over the 1990s.

Detailed results can be obtained from the author.

Very similar results were also found using real broad money (M2 + CDs).

Like output, these variables were normalized by dividing by potential output.

    Other Resources Citing This Publication