Chapter

9 Structural Change and Information Content of Financial Variables for Monetary Policy

Author(s):
Tamim Bayoumi, Guy Meredith, and Bijan Aghevli
Published Date:
June 1998
Share
  • ShareShare
Show Summary Details
Author(s)
Christopher Towe

A prerequisite for the successful implementation of monetary policy is the ability of policymakers to identify predictable relationships between instruments under their control and the ultimate objectives of policy. Since the 1970s, however, it has become increasingly apparent that these relationships can be disrupted by structural changes, including financial market innovation and deregulation.

For example, during the 1970s and early 1980s, many central banks maintained explicit or implicit monetary targets as a way of disciplining their actions and enhancing the credibility of inflation objectives. This approach was adopted principally in light of evidence suggesting that the aggregates could be reliably controlled by changes in interest rates, and that the aggregates were, in turn, reliably linked to output and inflation. However, in many cases, these targets were abandoned during the 1980s as it became apparent that deregulation and innovation in the money markets had caused large and unpredictable shifts in the links between the instruments, targets, and longer-term objectives of policy.

The breakdown of these historical relationships led a number of central banks to abandon explicit monetary targets in favor of an “eclectic” approach to policy. This approach involved the use of a range of indicators and information variables—including interest rates, yield curve spreads, inflation expectations, commodity prices, and exchange rates—to gauge whether policy is tight or loose relative to policymakers’ longer-term objectives.

However, the absence of clear-cut and reliable relationships between these indicator variables and the macroeconomy has raised concern about the ability of central banks to achieve their longer-term policy objectives. It has also made it more difficult for policymakers to credibly establish their commitment to these goals. To address this latter concern, a number of central banks have sought to improve the credibility of their policies by adopting explicit targets for their policy objective—the inflation rate.1

The effect of structural change on the usefulness of financial variables for guiding monetary policy has been demonstrated for a number of countries. For the case of the United States, Friedman and Kuttner (1992) demonstrate that the monetary and credit aggregates appear to contain significant information about future real activity and inflation prior to the 1980s, but that this relationship is not significant when post-1980 data are considered. Instead, interest rates (and interest rate spreads) are shown to contain significant explanatory power, which they argue highlights the importance of focusing on interest rates rather than the monetary aggregates as indicators of future price and output developments.

Attempts to investigate this issue for Japan have yielded results considerably less definitive and, unlike the United States, generally have not suggested as strong a role for interest rates in more recent periods. For example, both Ueda (1993) and Moreno and Kim (1993) find that the monetary aggregates contain important leading information even in the latter historical period, while the role of interest rates in predicting economic activity is ambiguous.

The question of whether financial variables can act as leading indicators for aggregate activity and prices is particularly relevant for Japan given the volatility of Japanese economic conditions during the past decade. In particular, the bubble phenomenon of the late 1980s, and the relatively sluggish recovery of the economy during the past several years, has led some analysts to question whether the usual indicators were providing the appropriate signals for monetary policy.

For example, Schinasi and Hargraves (1993) argue that financial market deregulation distorted the usual transmission mechanisms for monetary policy and contributed to overly stimulative monetary policies and excessive asset price inflation during the 1980s in many countries, including Japan. Lipworth and Meredith (chapter 11, this volume) also suggest that interest rates provided a misleading picture of the stance of policy during the late 1980s and early 1990s. They demonstrate that an index of “financial conditions,” which includes stock market prices and the exchange rate, indicates that policy was considerably easier in the late 1980s and tighter in the 1990s than suggested by interest rate developments.

The possibility that the information content of financial variables has not been stable is particularly relevant in view of the fact that deregulation in Japanese financial markets is a relatively recent phenomenon, and further substantial deregulation is being contemplated.2 The absence of a stable relationship between the indicators and the objectives of monetary policy may also be relevant in the context of the recent legislative initiative to increase the Bank of Japan’s independence, transparency, and accountability.3 The debate over this legislation tended to be focused on administrative issues, such as the composition of the Policy Board and whether its minutes would be published. While these issues are important, transparency and accountability also require that the central bank has the means to communicate in a credible fashion both its longer-run objectives and its current policy stance. In the latter case, this requires that the central bank have at its disposal reliable policy indicators.

A complete resolution of these issues is beyond the scope of this paper. Instead, the objective here is to assess the narrower issue of whether financial variables (including monetary aggregates, interest rates, and exchange rates) can be identified that contain information regarding the future course of GDP growth and inflation. Next, the paper examines whether these relationships are stable and, therefore, could be exploited by monetary policymakers.

The next section briefly discusses what is usually meant by an indicator, a target, and an instrument of monetary policy. The recent history of monetary policy implementation in Japan and the extent to which it has been affected by structural reform in the Japanese financial sector is then reviewed. The section that follows presents evidence to suggest that while there have been periods during which financial variables have provided useful information about future output and price developments, these relationships do not appear to have been stable. The last section contains concluding remarks.

Targets and Instrument of Monetary Policy

To discuss the role of financial variables as guides for monetary policy, it is helpful to clarify the distinction between targets, instruments, and indicators of policy.4 A monetary policy target is usually understood to be a proximate objective of policymakers, rather than the ultimate goal policy (such as price stability). Madigan (1994) also distinguishes between operating targets, which central banks aim for in the shorter run between policy decisions (such as a short-term interest rate) and intermediate targets (such as the money supply), which are intermediate between the operating target and the ultimate objective of policy.

Policy instruments are the variables that the central bank has under its direct control, whereas an indicator is a variable that is not necessarily under the control of the central bank, but provides a reliable signal of the stance of policy. In recent years, central banks have identified a large range of variables that meet this criterion, including the slope of the yield curve, commodity prices, and measures of real interest rates and inflation expectations. Many central banks have avoided explicitly defining the indicators used by policymakers, but other countries have adopted a more objective approach. For example, in Canada, a monetary conditions index (MCI)—the weighted sum of changes in the nominal interest rate and the exchange rate—has been explicitly defined by the central bank to identify the stance of policy.

Monetary policy indicators—including MCIs—can be viewed in different ways. Lipworth and Meredith (1996) describe an MCI for Japan, which is interpreted as an indicator of financial conditions rather than a measure of the stance of policy. Specifically, the index is based on multipliers derived from a structural model of the Japanese economy, and represents the net effect on aggregate demand of movements in the financial variables that are affected both by monetary policy and other market forces.

Information variables are similar to policy indicators in that they also exhibit a useful relationship with the objectives of policy. However, information variables are considered to be distinct from policy indicators in that their relationship to policy objectives is comparatively unpredictable, and dependent on economic and other circumstances. This is partly because the choice of which variable to classify as an information variable is usually based on evidence of a statistical relationship between the variables and measures of aggregate activity. It is also partly because information variables are usually used to predict shorter-term changes in economic activity and prices. As a result, while such variables can provide information about the overall stance of policy and the future course of the macroeconomy, they are considered to be less reliable predictors of future activity than policy indicators. The range of information variables used by central banks is often large and depends on circumstances.

Monetary Policy Implementation in Japan

The objectives of the Bank of Japan are defined relatively broadly. The Bank of Japan Law states that “The Bank of Japan has for its object the regulation of the currency, the control and facilitation of credit and finance, …, in order that the general economic activities of the nation be enhanced.” A number of authors have suggested (e.g., Batten and others, 1990) that the focus of policy has evolved over time, particularly since the move to a floating exchange rate and the first oil price shock in the early 1970s. For example, the emphasis appeared to shift from fostering growth during the time of the first oil price shock to one of containing inflation after the second oil price shock. During the 1980s, a period of rapid asset price inflation, exchange rate stability appeared to take on a greater priority. However, the low inflation rate in Japan since the late 1970s suggests a strong commitment to price stability. Indeed, Governor Matsushita stated recently that “the Bank [of Japan] attaches the utmost importance to realizing price stability.”5

The Bank of Japan’s approach to achieving its policy objectives is often described by observers as “eclectic” and “pragmatic.” This was confirmed by Governor Matsushita, who stated that “The Bank of Japan has never adopted stringent targeting or specific rules to date, but has consistently emphasized the judgment of the overall economic conditions,” Thus, in setting policy, the Bank of Japan takes into account a range of factors including price developments, economic activity, money supply, exchange rate, money market rates, and international considerations.

Reflecting this eclectic approach, the Bank of Japan has never adopted strict targets for the monetary aggregates. Nonetheless, monetary and financial aggregates have played an important role in the conduct of monetary policy. From 1960 to the early 1970s, a focus of Bank of Japan policy was lending by financial institutions to the nonbank sector. The instruments for achieving its objectives for lending growth included the interest rate in the interbank market (described below) and direct administrative control of bank loans through the use of “window guidance.” The latter involved informal instructions to individual financial institutions regarding the growth of their loans to the private sector.

Adherence to these guidelines was ensured by the fact that the Bank of Japan controlled banks’ access to the discount window, from which the Bank of Japan provided credit at below market rates.6 The use of window guidance for policy purposes waned in the 1980s and was abolished in June 1991. The decision to eliminate the use of direct credit controls as an instrument of policy was, in large part, the result of financial market liberalization. In particular, the private sector’s access to credit from financial institutions that were not subject to “guidance” began to increase, so that the impact on overall credit conditions of measures that only affected the banks began to diminish.7

Following the first oil price shock, the Bank of Japan explicitly shifted the focus of policy to M2 (subsequently to M2+CDs—broad money and certificate of deposits), and quarterly projections of the money supply began to be published in 1978.8 These quarterly projections were not strict monetary targets, and a number of observers have noted that there is limited evidence to suggest that the stance of monetary policy was altered in response to differences between the actual and projected levels of the money supply.9 Indeed, with the financial market liberalization and innovation that began in the mid-1980s, monetary aggregates became more volatile and thus less useful as leading indicators. As a result, the Bank of Japan further downgraded monetary aggregates as a guide for policy. Nonetheless, considerable attention continues to be paid to monetary aggregates, and the Bank of Japan’s statements that have accompanied interest rate cuts during 1995 referred to the sluggishness in money growth as a factor influencing interest rate policy.

Despite these changes, the Bank of Japan has consistently laid a strong emphasis on short-term interest rates as a day-to-day focus of monetary policy.10 In terms of the discussion above, the level of short-term interest rates has been the operating target for policy. However, the instruments that the Bank of Japan used to achieve its short-term interest rate objectives have evolved considerably in response to structural change and liberalization in the financial sector.

Up until late 1988, moral suasion was one of the principal means by which the Bank of Japan’s interest rate objectives were achieved, mainly owing to the relatively small size of the money market and the lack of a developed market for short-term instruments.11 As a result, during the period up to 1988, the overnight rate tended to be set directly by the Bank of Japan (Ueda, 1993, p.18). For example, under the tatene system (September 1967 to April 1979), the Bank of Japan would meet with a representative call loan dealer after the close of each day’s market and discuss the call rate for the next day, which would be announced at the opening of the following day. Under the kehaichi system (May 1979 to October 1988), the overnight rate was set daily by the Bank of Japan and was subsequently announced by the dealers. More public signals of the Bank of Japan’s interest rate objectives were (and continue to be) conveyed by changes in the official discount rate.

Substantial financial market deregulation and innovation occurred during the 1980s and early 1990s.12 During the mid- and late-1980s, the markets for uncollateralized call transactions, treasury bills, and commercial paper were developed. During the same period, restrictions on the range of financial institutions that were eligible to participate in the money market began to be lifted. In addition, the market for CDs was liberalized considerably; minimum denomination requirements were relaxed; and the range of CD maturities that could be offered was widened to include the shorter and longer terms. The interbank market was formally deregulated in 1988. Regulations on the level of interest rates, which had been linked to the official discount rate, also were gradually lifted, and by October 1994, the process of liberalizing deposit rates was completed.

As a result of these changes, the Bank of Japan’s operating procedures have evolved significantly from an emphasis on indirect methods for achieving its interest rate objectives toward a more market-based approach. For example, prior to mid-1995, the official discount rate (ODR) was customarily set below the overnight call rate. As a result, the Bank of Japan ensured an excess demand for its lending, and the principal instrument for affecting monetary conditions was through discretionary changes in the amount of its loans to the banking sector. However, in July 1995, the Bank of Japan announced a reduction in market interest rates to a level that was lower than the ODR, and since then, the overnight call rate has been at or just below the ODR. As a result, the demand for loans from the Bank of Japan has waned, and the role of Bank of Japan lending as a policy instrument has been virtually eliminated.

Significant changes in the stance of policy are still validated by (usually infrequent) changes to the ODR. However, this instrument is now mainly viewed as a signaling device. Under the new system, the Bank of Japan achieves its day-to-day interest rate objectives through transactions in treasury bills, financing bills (short-term government bills), cover bills (short-term securities issued by financial institutions), and commercial paper. Bank of Japan transactions in these markets can take the form of outright sales and purchases or can take the form of overnight sales and repurchase agreements. These transactions affect interest rates directly, but also have an indirect effect on money markets by adding or withdrawing reserves from the banking sector relative to reserve requirements.13

Structural Change and the Role of Information Variables

Despite the magnitude of the structural and institutional changes that have occurred in Japanese financial markets, the effect of these changes on the usefulness of financial variables as a guide for policy appears not to have been studied extensively.14 Moreover, as noted above, the studies that have considered the implications of structural change appear to have yielded conflicting results. Most important, and unlike the case of the United States, these studies do not find an increased role for interest rates as a leading information variable in recent years.

For example, Ueda (1993) finds that the inclusion of the 1980s in bivariate vector autoregressions (VARs) does not significantly diminish the role of the monetary aggregates in explaining industrial production, but that interest rates appear to be less significant if the latter period is included. Moreno and Kim (1993) find that M2+CDs rather than interest rates provides a useful indicator for monetary policy for 1960-80, but that in 1980-92, the usefulness of monetary aggregates as a leading indicator is substantially diminished. However, unlike the results for the United States, they find no corresponding improvement in the power of interest rates to predict real activity.

In the discussion below, the extent to which financial variables provide useful leading information for monetary policy, and whether these relationships have undergone structural changes, is reexamined. The analysis is conducted first using the type of Granger causality tests used by a number of previous authors, supplemented by the consideration of more recent data. These results are then used to develop a time-varying parameter model for forecasting real GDP growth.

Data Issues

Before examining the information content of financial variables, it is important to consider whether the data are stationary, since this will affect the statistical techniques that can be used. The evidence regarding the stationarity of Japanese macroeconomic data is mixed. Soejima (1995) notes that recent studies of Japanese postwar GNP data have suggested that the hypothesis of a unit root can be rejected if a break in the data in the early 1970s is considered. He extends these results to a broader range of economic variables, including real private consumption, industrial production, M2+CDs, and the call rate, and confirms that the unit root hypothesis can be rejected if the data are assumed to follow a trend with a break somewhere within 1971-73. However, he acknowledges that these results are contradicted because many of the series exhibit signs of a unit root when examined over 1975—93.15

To address this issue, augmented Dickey-Fuller (ADF) tests were conducted quarterly for a number of macroeconomic variables for 1960-96 and 1975-96 (for the sake of brevity, the results are not reported). While not fully conclusive, these tests suggested that the measures of real aggregate economic activity, prices, money supply, and the exchange rate used below were nonstationary in levels but were first difference stationary.16 The interest rate appeared to be stationary in the larger sample, but exhibited signs of nonstationarity in the shorter time period. Thus, for the purpose of the analysis below, it was assumed that the data were integrated of order one, except the interest rate data, which were assumed to be stationary in levels.17

Bivariate Causality Tests

Following the approach used by Friedman and Kuttner (1992) and others, the information content of financial market variables for measures of aggregate economic activity was examined using Granger causality tests.18 This technique enables the identification of variables that provided significant information for predicting the future course of aggregate activity and inflation and, therefore, that could be useful information variables for monetary policy.

In particular, bivariate vector autoregressions of the following form were estimated using quarterly data:19

where Yt represents the particular index of aggregate economic activity or prices of interest to monetary authorities and Pt represents a financial variable that could be considered a possible information variable. The measures of real activity considered below included real GDP, real total domestic demand, and real private domestic demand. The aggregate price indices included the CPI, the core CPI, the GDP deflator, and the consumption deflator. Candidates for status as a leading financial indicator included the call rate, the Gensaki rate, the three-month CD rate, the 10—year bond rate, M2+CDs, M2, M1, stock prices, the yen/dollar exchange rate, the nominal effective exchange rate, and the real effective exchange rate (the interest rate and monetary variables defined in “real” terms are also considered).

A potential drawback to this approach is that it does not take into account the contemporaneous correlations between variables, that is, that the financial variable may affect output or inflation in the same period. This possibility was not addressed because the assumptions required to identify the direction of the contemporaneous causality are somewhat arbitrary, and the effect of financial variables (and monetary policy in general) is usually thought to be felt on the macroeconomy only with a lag.20 Moreover, an examination of the residuals from the VARs suggested that the cross-equation correlations were usually small and insignificant from zero.21

Tables 9.1 and 9.2 summarize results of the bivariate causality tests described above for the 1962:Q1-1996:Q2 period. The latter table includes a dummy variable for the post-1974 period, to account for the structural shift identified by Soejima (1996).22 These results strongly suggest that both the narrow and the broader monetary aggregates, defined either in nominal or real terms, acted as useful leading indicators for aggregate economic activity (i.e., the p- values are well below 5 percent). The evidence that monetary aggregates reliably predict price inflation is less strong, particularly when the regressions included a post-1974 dummy, perhaps reflecting the dominant role of commodity price shocks on inflation during this period. However, a significant leading relationship (at the 5 percent level) was found between the growth of M2+CDs (or the growth of M2) and the core inflation rate.

Table 9.1.Granger Causality Tests, 1962–961(P-values)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.000*0.001*0.001*0.005*0.001*0.003*0.008*
M2 (first difference)0.000*0.001*0.001*0.006*0.001*0.006*0.016*
M1 (first difference)0.004*0.002*0.005*0.1020.1390.0670.080
Real monetary aggregates
M2+CDs (first difference)0.000*0.000*0.000*0.005*0.001*0.001*0.011*
M2 (first difference)0.000*0.000*0.000*0.006*0.001*0.003*0.022*
M1 (first difference)0.008*0.001*0.001*0.1020.1380.2420.175
Interest rates
Call rate0.1730.0850.024*0.0720.1050.0520.118
Gensaki rate0.004*0.002*0.000*0.000*0.000*0.000*0.004*
Ten-year rate0.8250.2700.3740.0720.1760.1090.031*
Yield curve spread0.009*0.017*0.004*0.001*0.000*0.000*0.151
Call rate (first difference)0.1810.1520.028*0.049*0.3750.016*0.187
Gensaki rate (first difference)0.010*0.016*0.024*0.000*0.001*0.000*0.011*
Ten-year rate (first difference)0.4560.2380.4240.1200.5120.3560.152
Real interest rates
Call rate0.5640.9540.6780.031*0.7790.015*0.289
Gensaki rate0.7400.7280.5480.002*0.6740.4490.015*
Ten-year rate0.1120.3400.3630.6070.3910.1070.679
Call rate (first difference)0.2680.0930.1260.026*0.7930.1570.315
Gensaki rate (first difference)0.8350.8130.4860.003*0.6900.0760.959
Ten-year rate (first difference)0.1050.2630.4760.4530.6560.000*0.538

P-values are the result of bivariate Granger causality tests, and indicate the probability that the interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity inflation. Asterisks denote p-values equal to or less than 5 percent.

P-values are the result of bivariate Granger causality tests, and indicate the probability that the interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity inflation. Asterisks denote p-values equal to or less than 5 percent.

Table 9.2.Granger Causality Tests, With Dummy Variable, 1962–961(P-values)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.009*0.007*0.001*0.0690.039*0.1060.144
M2 (first difference)0.013*0.008*0.003*0.0870.050*0.1810.249
Ml (first difference)0.0640.017*0.020*0.5740.8230.5680.462
Real monetary aggregates
M2+CDs (first difference)0.000*0.000*0.000*0.0690.031*0.008*0.080
M2 (first difference)0.000*0.000*0.000*0.0870.045*0.019*0.139
Ml (first difference)0.014*0.003*0.002*0.5740.8520.4830.774
Interest rates
Call rate0.0690.012*0.005*0.0950.2780.0620.198
Gensaki rate0.002*0.001*0.000*0.000*0.007*0.000*0.011*
Ten-year rate0.9930.1950.3470.016*0.030*0.024*0.003*
Yield curve spread0.005*0.012*0.002*0.001*0.000*0.000*0.157
Call rate (first difference)0.0820.040*0.0560.0540.3360.017*0.222
Gensaki rate (first difference)0.006*0.013*0.024*0.000*0.001*0.000*0.009*
Ten-year rate (first difference)0.3740.2020.4110.1100.5340.4030.380
Real interest rates
Call rate0.0910.1990.1670.014*0.5200.006*0.110
Gensaki rate0.3430.4510.4310.000*0.3280.2960.341
Ten-year rate0.041*0.1540.1530.4240.3770.1520.827
Call rate (first difference)0.1470.4640.4490.017*0.5160.1300.207
Gensaki rate (first difference.)0.6940.8570.4480.000*0.000*0.004*0.558
Ten-year rate (first difference)0.0580.1960.3970.4100.5840.001*0.612

P-values are the result of bivariate Granger causality tests, and indicate the probability that the Interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity inflation. Asterisks denote p-values equal to or less than 5 percent.

P-values are the result of bivariate Granger causality tests, and indicate the probability that the Interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity inflation. Asterisks denote p-values equal to or less than 5 percent.

Strong evidence that nominal short-term interest rates lead economic activity also was found. The p-values for both the Gensaki rate and the call rate were below the 5 percent level for most of the output variables examined, regardless of whether interest rates were introduced in levels or changes, or whether the regression included a dummy. Interestingly, while the 10-year bond rate was generally not found to lead output, the yield spread (the difference between the three-month and the 10-year rates) was significant.23

Evidence also existed for a causal relationship running from interest rates to prices. In particular, the Gensaki rate (both in levels and in first differences), the yield spread, and the 10-year bond rate appeared to predict a significant portion of the variability of the price indices. However, the confidence that could be placed in this result was diminished because coefficient estimates indicated a positive correlation between inflation and lagged interest rates. This suggests that the information contained in the interest rate variables for inflation was chiefly with regard to inflation expectations, and this possibility was supported by the fact that the real interest rate series were considerably less significant in the inflation equations. The same positive correlation was found even when VARs were rerun using the change in the inflation rate as the dependent variable.

To gauge the stability of these results, and to examine the possibility that structural change in the financial sector affected the role of the financial variables as leading indicators for monetary policy, the tests were repeated for the 1980:Q1-1996:Q2 period (Table 9.3). Truncating the sample excluded the 1970s, the period of the first oil price shock and the initial years of the flexible exchange rate regime. Moreover, since the process of financial sector reform and liberalization began in 1979, with the abolition of the call rate system and the establishment of the CD market, the truncated sample would be expected to be more representative of the modern financial system in Japan.

Table 9.3.Granger Causality Tests, 1980–961(P-values)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0002*0.0550.005*0.3650.2910.8950.714
M2 (first difference)0.003*0.0780.011*0.2830.1720.9350.695
M1 (first difference”)0.4940.0740.031*0.2360.6830.4490.390
Real monetary aggregates
M2+CDs (first difference)0.007*0.011*0.002*0.3650.4430.6150.806
M2 (first difference)0.016*0.024*0.006*0.2830.2920.5900.758
Ml (first difference)0.7110.1110.0710.2360.5800.0770.255
Interest rates
Call rate0.7630.4530.5410.027*0.020*0.000*0.129
Three-month CD rate0.7660.6570.7850.035*0.0620.1890.403
Ten-year rate0.6990.2810.7780.001*0.001*0.000*0.003*
Yield curve spread0.1480.1670.3410.9170.050*0.049*0.803
Call rate (first difference)0.7340.4650.6760.026*0.1530.013*0.325
Three-month CD rate (first difference)0.6480.8100.8660.0580.4110.7680.751
Ten-year rate (first difference)0.9290.6140.9810.018*0.2900.7350.430
Real interest rates
Call rate0.7380.5340.6230.155*0.0930.1730.230
Three-month CD rate0.2440.8680.5830.041*0.086*0.2750.356
Ten-year rate0.7340.8490.3720.332*0.831*0.1510.753
Call rate (first difference)0.045*0.009*0.005*0.005*0.111*0.385*0.094*
Three-month CD rate (first difference)0.048*0.009*0.013*0.038*0.076*0.672*0.054*
Ten-year rate (first difference)0.1530.045*0.043*0.502*0343*0.296*0.216*
Exchange rates-stock prices
Yen/U.S. dollar (first difference)0.2620.8360.4040.1710.5310.6190.309
Nominal effective (first difference)0.0910.3540.1610.0990.6750.4770.728
Real effective (first difference)0.2960.7710.3980.0860.3380.1020.495
Nikkei 225 (first difference)0.2770.1650.1820.6770.5600.03310.097

P-values are the result of bivariate Granger causality tests, and indicate the probability that the interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity/inflation. Asterisks denote p-values equal to or less than 5 percent.

P-values are the result of bivariate Granger causality tests, and indicate the probability that the interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity/inflation. Asterisks denote p-values equal to or less than 5 percent.

As discussed above, the expectation is that the financial sector reforms undertaken during this period would have weakened the relationship between monetary aggregates and measures of economic activity, and strengthened the role of interest rates, as has been found in the United States. Surprisingly, however, the opposite result was evident. In particular, monetary aggregates (defined either in nominal or real terms) remained a significant leading indicator of real economic activity, but no longer led inflation. Conversely, nominal interest rates and the yield curve spread no longer contained significant leading information about output, but appeared to lead inflation. The real interest rate series, when defined in first differences, were the most consistent-leading variables for real activity and inflation, albeit still with the wrong sign in the case of inflation.

The role of exchange rates in predicting real economic activity and prices was also considered. However, neither the nominal nor the real exchange rate was found to be a significant explanatory variable for the measures of real activity. The only significant relationship was between the real effective exchange rate index and the CPI and the consumption deflators.24 The hypothesis that stock prices (the Nikkei 225 index) could provide significant leading information about aggregate activity was also rejected.

An explanation for these results could be that they were affected by the collapse of the bubble economy during the 1990s. These events could have distorted the usual transmission mechanism for monetary policy and affected the role of financial variables as leading indicators. To consider this possibility, the tests were rerun over the 1980:Q1-1990:Q3 period (stock market prices peaked in mid-1990) to exclude the collapse of the asset-price bubble (Table 9.4).

Table 9.4.Granger Causality Tests, 1980–901(P-values)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.5620.5810.3440.9950.9581.0000.536
M2 (first difference)0.3310.3650.3340.8720.9660.7790.258
M1 (first difference)0.1410.044*0.011*0.3400.5540.6300.882
Real monetary aggregates
M2+CDs (first difference)0.6080.3250.2130.9950.6730.027*0.497
M2 (first difference)0.4200.2050.2070.8720.9430.027*0.248
Ml (first difference)0.1590.0660.032*0.3400.4790.7560.944
Interest rates
Call rate0.011*0.018*0.002*0.1400.1980.0020.098
Three-month CD rate0.1530.0930.0640.1460.3410.0270.766
Ten-year rate0.012*0.000*0.000*0.0680.034*0.009*0.046*
Yield curve spread0.2760.5680.5380.9050.9740.1620.482
Call rate (First difference)0.2790.2270.3080.0840.1520.0570.576
Three-month CD rate (first difference)0.7370.8590.0860.0680.4390.6900.711
Ten-year rate (first difference)0.7220.7500.0600.0250.2750.3410.003
Real interest rates
Call rate0026*0.0800.0550.9240.8740.2760.865
Three-month CD rate0.1290.2920.3280.1470.6730.2180.628
Ten-year rate0.6440.8880.6500.9820.3800.037*0.764
Call rate (first difference)0.005*0.004*0.004*0.046*0.0720.9530.412
Three-month CD rate (first difference)0.003*0.002*0.006*0.1790.036*0.8560.787
Ten-year rate (first difference)0.2040.0920.0900.041*0.2600.2140.378
Exchange rates/stock prices
Yen/U.5. dollar (first difference)0.5160.4740.7860.0550.1360.2180.190
Nominal effective (first difference)0.2500.9930.3840.0730.2260.4550.214
Real effective (first difference)0.5200.5980.6590.027*0.0990.2090.093
Nikkei 225 (first difference)0.1070.040*0.064-0.8620.4630.5160.066

P-values are the result of bivariate Granger causality tests, and indicate the probability that the interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity/inflation. Asterisks denote p-values equal to or less than 5 percent.

P-values are the result of bivariate Granger causality tests, and indicate the probability that the interest rates, monetary aggregates, and exchange rates do not cause the indices of real activity/inflation. Asterisks denote p-values equal to or less than 5 percent.

The results for the 1980s conformed somewhat more closely to the maintained hypothesis that the role of monetary aggregates would have diminished compared with that in the 1960s and 1970s. In particular, M2 and M2+CDs no longer acted as a leading indicator for economic activity or prices (Ml was marginally significant). Also, during the abbreviated sample, the level of (nominal) interest rates was consistently significant in the output equations. As before, the change in real short-term interest rates was a significant leading indicator for real activity. Exchange rate indices did not appear to have a significant role as leading indicators for either the price or output variables.

Single Equation Estimates

To provide more concrete information on the information content of financial variables for forecasting aggregate activity, a simple forecasting equation for real GDP growth was estimated for 1980:Q1-1996:Q2, using recursive least squares:

where Δ represents first differences, yt is real GDP, it is the three-month CD rate, mt is M2+CDs, and et is the nominal effective exchange rate. The lag length was based on the results of the bivariate VARs estimated above, which used the Schwarz information criteria.

The path of the coefficient estimates, shown in Figure 9.1, confirmed the instability of information content of the financial variables, particularly between the 1980s and the 1990s. In particular, the forecasting ability of the three-month CD rate appears to have steadily increased during the 1980s (i.e., the coefficient, and its t-statistic, increased in absolute value). However, after 1990, the interest rate coefficient returned to a level that was insignificant from zero. By contrast, the coefficient on the monetary aggregate was insignificantly different from zero during the 1980s, but was significantly different from zero during the 1990s. While the exchange rate coefficient had the correct sign, it was never statistically different from zero.25

Figure 9.1.Recursive Coefficient Estimates: Real GDP Equation With Nominal Financial Variables

Source: IMF staff estimates.

Since the causality tests above suggested a role for the financial variables defined in real terms, the equation above was reestimated to include a two-period lag of the real three-month CD rate (in first differences), and a one-period lag of the growth of real M2+CDs and the rate of change of the real effective exchange rate on the right-hand side. The real interest rate was included in first difference form principally so as to enable the comparison of the results with the weights in the MCI.

Recursive coefficient estimates are illustrated in Figure 9.2, and are similar to those for nominal financial variables. In particular, they confirmed that the information content of short-term interest rates appeared to increase during the 1980s, but that this process was interrupted in the early 1990s. These latter estimates also suggested that the information content of the monetary aggregate and the exchange rate (this time defined in real terms) was minimal, but that monetary aggregates contained significant information content in the 1990s.

Figure 9.2.Recursive Coefficient Estimates: Real GDP Equation With Real Financial Variables

Source: IMF staff estimates.

To explore this instability further, simple forecast equations for real GDP growth were estimated that included a break in 1990:Q3. The break point was roughly coincident with the collapse in equity prices that began in mid-1990, but was also suggested by Chow tests, which showed a break in that quarter at the 90 percent confidence level. The first equations included the same nominal financial indicators described above:

but also tested explicitly for a structural break in 1990 by the inclusion of d90, a dummy variable that was set equal to zero up to 1990:Q3 and equal to unity thereafter. The second regression equation included the same financial variables expressed in real terms:

where rit is the real interest rate, rmt is the change in the log of real M2+CDs, and ret is the change in the log of the real effective exchange rate.

Estimates for the nominal variables are presented in Table 9.5. The results are strongly suggestive of a structural change in the forecasting ability of the three financial variables for real output growth. In particular, the monetary aggregate did not contain significant leading information regarding GDP growth during the 1980s, but was a significant leading indicator after this date. By contrast, the three-month CD rate was a significant leading indicator during the 1980s, but during the 1990s, the interest rate was no longer significant. (In particular, the hypothesis that the sum of the coefficients on the interest rate (α26) equaled zero could not be rejected at the 95 percent confidence level.) The nominal effective exchange rate was not significant during the 1980s but became significant in the 1990s. Appendix II illustrates the results of a more general specification of equation (3), in which coefficients were allowed to vary continuously over time, estimated using a Kalman filter. The results were roughly similar.

Table 9.5.GDP Forecast Equation—Nominal Financial Variables
Coefficient EstimatesDiagnostic Statistics
α0α1α2α3α4α5α6α7α8R2DH
Unconstrained0.015-0.278-0.1120.190-0.019-0.0160.1660.797-0.0570.370.05
(2.35)(2.23)(0.93)(0.93)(0.83)(2.32)(2.13)(2.30)(1-53)
Constrained0.020-0.258-0.1250.1790.972-0.0750.350.14
(5.08)(2.08)(2.35)(2.31)(3.37)(2.53)
Note: Estimated using ordinary least squares; t-statistics are in parentheses; DH refers to Durbin’s H statistic, which has a t-distribution.
Note: Estimated using ordinary least squares; t-statistics are in parentheses; DH refers to Durbin’s H statistic, which has a t-distribution.

The information content of financial variables defined in real terms was somewhat more stable (see Table 9.6). For example, real money growth and the real interest rate were significant in both subperiods. Nonetheless, the coefficient on the monetary aggregate increased in significance in the latter period, suggesting a rise in its information content. Unlike the nominal regression, the coefficient for the real interest rate remained significant in the latter period and, indeed, its significance increased. The role of the real effective exchange rate also was found to have increased in the latter period.

Table 9.6.GDP Forecast Equation—Real Financial Variables
Coefficient EstimatesDiagnostic Statistics
β0β1β2β3β4β5β6β7β8R2DH
Unconstrained0.007-0.269-0.4050.336-0.021-0.009-1.3290.766-0.0250.510.08
(3.81)(2.38)(2.89)(3.09)(0.98)(3-26)(3.02)(2.52)(0.83)
Constrained0.007-0.275-0.3280.316-0.009-1.4100.791-0.0460.510.19
(4.01)(2.43)(2.82)(2.96)(3.38)(3.26)(2.61)(2.15)
Note: Estimated using ordinary least squares; t-statistics are in parentheses; DH refers to Durbin’s H statistic. In these regressions, the dummy variable was assumed to equal one after 1991:Q3.
Note: Estimated using ordinary least squares; t-statistics are in parentheses; DH refers to Durbin’s H statistic. In these regressions, the dummy variable was assumed to equal one after 1991:Q3.

A number of aspects of these results are worth highlighting. First, the recursive estimates shown in Figures 9.1 and 9.2, and the simple regression results reported in the tabulations above, yielded somewhat surprising results regarding the information content of the monetary aggregate. In particular, the increased significance of M2+CDs in the 1990s, and the concurrent decline in the significance of the nominal interest rate variable, is striking, since the deregulation that took place would have been expected to have the opposite effect.

One explanation for the increased significance of M2+CDs is that the banking sector’s attempts to meet BIS capital-adequacy requirements, as well as the effect of the collapse of equity prices on banks’ balance sheets, led to nonprice constraints on banks’ ability to supply credit to the private sector, particularly during the early 1990s.26 For example, domestic credit growth and the growth of M2+CDs fell sharply after mid-1990 from more than 10 percent to near zero in 1993. Previous research in this area suggests that this phenomenon may have been important, particularly in the early 1990s.27 For example, the Tankan survey’s index of banks’ willingness to lend fell sharply in 1990, and remained at a low level until 1994. Similarly, spreads between the short-term loan rate and the Gensaki rate rose in 1991 and have remained high. While statistical analysis has suggested that the period of “abnormal” spreads was somewhat shorter lived than these developments might suggest, there appears to be evidence of credit rationing during the early 1990s.

A second noteworthy feature of the estimates was the reduced role of nominal interest rates for forecasting future economic activity during the 1990s. That nominal rates were brought to levels that were close to zero during this period, and that real interest rates fell more slowly than nominal rates, may help explain this result. Indeed, estimates suggest that the importance of real interest rates for forecasting real GDP increased after 1991 (see Figure 9.2 and Table 9.5). The increased role of the exchange rate during the 1990s for forecasting real GDP developments is less easy to explain, but clearly suggests that the yen’s appreciation during this period weighed heavily on aggregate economic activity.

More generally, the estimates appear to suggest that the information content of financial variables is considerably greater when defined in real terms. This was evidenced by the greater explanatory power of the “real” regression, and the fact that the significance of the real variables appeared to be less unstable.

It is also noteworthy that coefficients on the change in the real interest rate and the percentage change in the real effective exchange rate in Figure 9.2 are similar to the weights used in the monetary conditions index calculated by Lipworth and Meredith (1996). While these results are not directly comparable (since, for example, the MCI is based on the simulation of an annual structural model), they provide an encouraging confirmation of the MCI weights. Indeed, when the equations described above were reestimated over 1980—96. excluding the monetary aggregate, a 10:1 ratio between the coefficient on the real interest rate and the coefficient on the real effective exchange rate was obtained—the same ratio as in the MCI.

Conclusions

The results presented above strongly suggest the existence of a structural break in the information content of financial variables for forecasting real activity and prices in Japan. In particular, and like the analysis performed for the United States, Granger-causality tests suggested that the role of monetary aggregates as a leading information variable fell during the 1980s, compared with the previous period, while the role of interest rates rose. This can be interpreted as evidence of the effect of deregulation in financial markets. However, unlike the case of the United States, this process was reversed during the 1990s, and the role of nominal interest rates for forecasting real activity diminished in the more recent period, while the role of monetary aggregates increased.

An explanation for the increased significance of the monetary aggregate is that attempts by the banking sector to meet BIS capital-adequacy requirements, and balance sheet problems in the wake of the collapse of the asset-price bubble, affected credit conditions by more than would be suggested by interest rate developments. The information content of nominal interest rates also may have diminished because of downward rigidities in nominal rates as they approached historical lows. This was confirmed by the fact that the significance of real interest rates for forecasting real GDP appeared to increase during the 1990s.

Before drawing any policy-related conclusions from these results, some important methodological caveats to the analysis should be acknowledged. In particular, Granger-causality tests are often criticized for their lack of an explicit structural framework. This lack of a structural framework is argued to limit their usefulness, and this criticism is even more salient in the case of simple, bivariate estimates. Moreover, while data were found to be nonstationary, explicit consideration was not given to the possibility that data were co-integrated—that is, exhibited a longer-run relationship in levels. Finally, the fact that some of the results were difficult to explain (e.g., the incorrect sign on interest rates in the inflation equations and the relatively short-lag structure in the output equations) reduces to some extent the degree to which broad conclusions can be drawn from the results.

Methodologies that addressed these shortcomings would be worth exploring in the future. For example, a richer empirical framework (e.g., a structural VAR or vector cointegrating system) would allow consideration of the interrelationships between the variables considered and the possibility that the instability was due to changes in the nature of shocks to the economy (e.g., aggregate supply versus aggregate demand shocks), or due to shifts in the Bank of Japan’s reaction function.28

Nonetheless, the results suggested that a single variable, or group of variables, may not yet exist that could be used as a consistent operational guide for monetary policy in Japan. To some extent, this confirms the appropriateness of the Bank of Japan’s “eclectic” and “pragmatic” approach to policy, which focuses on a range of indicators to guide policy.

However, the difficulty of obtaining a stable reduced-form relationship between the information variables and aggregate output and prices may also point to the advantage of a more structural approach to developing monetary indicators. For example, while strong conclusions regarding the MCI cannot be drawn from the results above, given the differences in approaches used, the results were encouragingly consistent with the interest rate and exchange rate weights in the index developed for Japan by Lipworth and Meredith (Chapter 11, this volume). Since these MCI-type indices rely on simulations of longer-run structural relationships, they may be a more useful and stable guide for policy than the shorter-run statistical correlations described above.

The results also have implications for the effort to increase the transparency and openness of monetary policy in Japan. As noted above, discussions have tended to focus on administrative issues. Less attention appears to have been paid to whether the Bank of Japan has reliable policy indicators at its disposal that would enable it to communicate credibly its current policy stance.

In particular, the fact that the information content of financial variables has been so unstable suggests that the burden on policymakers’ judgment in setting policy is large and possibly will intensify with the additional deregulation now being contemplated. The absence of reliable information variables also may make it difficult for the private sector to interpret the Bank of Japan’s policy stance and to gauge the extent to which the Bank of Japan is adhering to its stated policy intentions. At the same time, the instability of these relationships could reduce the ability of the Bank of Japan to communicate such intentions effectively. While the Bank of Japan’s efforts to improve transparency through the use of frequent public-statements by senior bank officials has been useful, there may be scope for increasing transparency further by making more explicit the indicators that guide policymakers.

Appendix I. Source of Data

This appendix describes the variables used in the analysis above. The data were obtained from databases compiled by Nikkei Telecom, Nomura Research Institute, and IMF staff (original data sources were Annual Report on National Accounts, Economic Planning Agency (EPA); Economic Statistics Monthly, Bank of Japan; and International Financial Statistics (IMF)).

Real growth indices: Real growth was defined as the log difference of quarterly GDP, total domestic demand, and total private domestic demand, in constant 1990 prices (seasonally adjusted).

Inflation indices: Inflation was defined as the log difference of the seasonally adjusted, quarterly average of the CPI, the core CPI (excluding fresh food, fuel, water, and light), the GDP deflator, and the private consumption deflator.

Monetary aggregates: Monetary growth rates were defined as the log difference of seasonally adjusted, end-quarter values of M2+CDs, Ml, and M2. Real monetary aggregates were constructed by deflating the nominal values by the quarterly average of the CPI. The core CPI was used for the single equation estimates.

Interest rates and stock price index: Interest rates included the call rate in Tokyo, the three-month repurchase rate (Gensaki rate), the three-month CD selling rate in Tokyo, and the five-year bond rate (interest-bearing bank debentures). The stock price index was the Nikkei 225 index. All series were quarterly averages. The real interest rate series were constructed by subtracting the CPI inflation rate during the previous four quarters.

Exchange rates: Exchange rate series used included the period averages of the yen per dollar rate, as well as IMF estimates of the nominal effective exchange rate and the real effective exchange rate (based on normalized unit labor costs).

Appendix II. Additional Results

Contemporaneous Correlations

To examine the contemporaneous correlations between the indicators of real growth and inflation and the financial indicators, the residuals from the bivariate VARs described in the section on structural change and the role of information variables were regressed on each other.29 The results are summarized in Tables 9.79.10, and suggest the following conclusions:

Table 9.7.Contemporaneous Correlations, 1962-961Coefficient (t-statistic)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.0210.0420.010-0.035-0.102-0.105-0.021
(0.209)(0.379)(0.088)(0.378)-(1.249)-(1.119)-(0.234)
M2 (first difference)-0.0170.001-0.001-0.029-0.094-0.079-0.022
-(0.189)(0.005)-(0.005)-(0.340)-(1.249)-(0.933)-(0.267)
M1 (first difference)0.0150.0160.0180.0070.0080.0050.003
(0.550)(0.530)(0.567)(0.284)(0.340)(0.191)(0.139)
Real monetary aggregates
M2+CDs (first difference)0.3350.3030.228-0.543-0.386-0.314-0.415
(3.375)(4.552)(3.103)-(13.144)-(9.566)-(5.579)-(8.700)
M2 (first difference)0.2280.2590.203-0.494-0.354-0.287-0.380
(3.701)(4.032)(2.888)-(11.864)(8.883)-(5.263)-(8.143)
M1 (first difference)0.0490.0590.056-0.072-0.043-0.048-0.063
(1.856)(2.124)(1.892)-(3.101)-(2.041)-(2.011)-(2.812)
Interest rates
Call rate0.0000.0000.0010.0040.0040.0020.003
-(0.058)-(0.345)(0.619)(3.501)(3.807)(1.373)(2.611)
Gensaki rate0.0040.0020.0000.0000.0000.0000.004
(0.004)(0.002)(0.000)(0.000)(0.000)(0.000)(0.004)
Ten-year rate0.000-0.001-0.0010.0050.0040.0020.004
(0.025)-(0.744)-(0.431)(3.215)(2.857)(1.665)(2.728)
Yield curve spread-0.002-0.002-0.0020.0040.0030.0020.004
-(2.175)-(2.325)-(1.681)(4.827)(5.109)(2.638)(4.304)
Call rate (first difference)0.0000.0000.0010.0040.0030.0010.003
-(0.156)(0.000)(0.785)(3.536)(3.195)(0.875)(2.705)
Gensaki rate (first difference)-0.001-0.0020.0000.0050.0040.0020.004
-(1.113)-(1.340)(0.364)(6.779)(6.456)(2.781)(5.129)
Ten-year rate (first difference)0.000-0.001-0.0010.0040.0030.0020.004
(0.029)-(0.617)-(0.349)(2.797)(2.498)(1.308)(2.502)
Real interest rates
Call rate0.0020.0020.002-0.007-0.004-0.003-0.004
(2.672)(3.433)(2.754)-(12.937)-(7.933)-(6.123)-(9.912)
Gensaki rate0.0020.0020.003-0.004-0.003-0.003-0.004
(2.122)(2.786)(3.449)-(5.422)-(3.470)-(4.538)-(6.177)
Ten-year rate0.0030.0030.003-0.007-0.005-0.003-0.005
(4.019)(3.839)(3.567)-(14.296)-(11.259)-(7.005)-(10.503)
Call rate (first difference)0.0020.0020.002-0.006-0.004-0.003-0.004
(3.149)(3.104)(2.708)-(12.274)-(7.927)-(5.924)-(10.049)
Gensaki rate (first difference)0.0010.0020.003-0.004-0.003-0.003-0.004
(1.698)(2.354)(3.092)-(4.863)-(3.418)-(4.372)-(4.881)
Ten-year rate (first difference)0.0030.0030.003-0.007-0.005-0.003-0.005
(4.325)(4.028)(3.513)-(14.029)-(11.409)-(6.389)-(9.925)

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

Table 9.8.Contemporaneous Correlations, VAR with Dummy Variable, 1962-961Coefficient (t-statistic)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.0150.0500.037-0.054-0.130-0.128-0.046
(0.150)(0.445)(0.308)-(0.567)-(1.642)-(1.329)-(0.505)
M2 (first difference)-0.0250.0050.021-0.047-0.124-0.102-0.047
-(0.272)(0.052)(0.195)-(0.547)-(1.690)-(1.175)-(0.572)
Ml (first difference)0.0040.0160.022-0.013-0.028-0.013-0.017
(0.145)(0.475)(0.605)-(0.449)-(1.272)-(0.477)-(0.653)
Real monetary aggregates
M2+CDs (first difference)0.3350.3160.222-0.550-0.380-0.340-0.426
(3.375)(4.976)(3.006)-(13.546)-(9.824)-(6.163)-(9.235)
M2 (first difference)0.2280.2700.198-0.503-0.352-0.313-0.394
(3.930)(4.394)(2.795)-(12.287)-(9.236)-(5.877)-(8.731)
Ml (first difference)0.0490.0650.063-0.107-0.074-0.083-0.095
(1.843)(2.259)(1.990)-(4.305)-(3.760)-(3.283)-(4.074)
Interest rates
Call rate0.000-0.0010.0010.0040.0040.0010.003
-(0.340)-(0.669)(0.445)(3.262)(3.739)(1.165)(2.356)
Gensaki rate0.0020.0010.0000.0000.0070.0000.011
(0.002)(0.001)(0.000)(0.000)(0.007)(0.000)(0.011)
Ten-year rate0.000-0.002-0.0010.0050.0030.0020.004
-(0.132)-(0.896)-(0.485)(3.059)(2.747)(1.481)(2.644)
Yield curve spread-0.002-0.003-0.0010.0040.0030.0020.003
(2.476)-(2.982)-(1.213)(4.735)(5.065)(2.656)(4.086)
Call rate (first difference)0.0000.0000.0020.0030.0030.0010.002
-(0.048)(0.024)(1.178)(3.390)(2.963)(0.763)(1.482)
Gensaki rate (first difference)-0.001-0.0020.0000.0040.0040.0020.003
-(1.185)-(1.393)(0.344)(6.281)(5.881)(2.363)(4.562)
Ten-year rate (first difference)0.000-0.001-0.0010.0040.0030.0010.003
-(0.123)-(0.725)-(0.385)(2.519)(2.233)(1.033)(2.207)
Real interest rates
Call rate0.0030.0030.003-0.006-0.004-0.003-0.004
(4.594)(4.524)(4.017)-(12.788)-(7.228)-(5.959)-(9.697)
Gensaki rate0.0020.0030.004-0.004-0.002-0.002-0.003
(2.894)(3.414)(3.968)-(4.689)-(2.866)-(3.706)-(4.226)
Ten-year rate0.0030.0040.003-0.007-0.005-0.003-0.005
(4.608)(4.355)(4.217)-(13.594)-(10.762)-(7.239)-(9.648)
Calf rate (first difference)0.0020.0020.002-0.006-0.004-0.003-0.004
(3.769)(3.877)(3.374)-(11.959)-(7.213)-(5.477)-(9.630)
Gensaki rate (first difference)0.0020.0020.003-0.004-0.002-0.002-0.003
(2.000)(2.506)(3.120)-(4.898)-(3.125)-(2.486)-(3.845)
Ten-year rate (first difference)0.0030.0040.003-0.007-0.005-0.003-0.005
(4.545)(4.174)(3.694)-(13.463)-(10.868)-(6.497)-(8.993)

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

Table 9.9.Contemporaneous Correlations, 1980-961Coefficient (t-statistic)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.1170.2310.277-0.057-0.094-0.0710.043
(0.997)(1.755)(1.997)-(0.776)-(1.880)-(0.602)(0.440)
M2 (first difference)0.0270.1000.187-0.076-0.098-0.064-0.004
(0.261)(0.859)(1.534)-(1.215)-(2.358)-(0.822)-(0.047)
Ml (first difference)0.0170.0490.044-0.033-0.047-0.074-0.068
(0.390)(1.029)(0.836)-0.161)-(2.431)-(2.338)-(0.898)
Real monetary aggregates
M2+CDs (first difference)0.3350.3160.342-0.279-0.184-0.131-0.153
(3.375)(3.142)(3.230)-(5.548)-(5.089)-(1.823)-(1.949)
M2 (first difference)0.0980.2110.268-0.237-0.159-0.125-0.144
(1.113)(2.270)(2.761)-(5.329)-(5.084)-(0.975)-(2.114)
Ml (first difference)0.0120.0460.036-0.077-0.064-0.090-0.098
(0.319)(1.122)(0.765)-(3.025)-(3.837)-(3.396)-(3.185)
Interest rates
Call rate0.0020.0020.0030.0030.0010.0010.002
(1.543)(1.068)(2.017)(3.298)(2.513)(1.721)(1.958)
Gensaki rate0.7660.6570.7850.035(0.062)(0.189)(0.403)
(0.766)(0.657)(0.785)(0.035)(0.062)(0.189)(0.403)
Ten-year rate0.0020.0010.0010.0010.0010.0010.002
(1.436)(0.468)(0.384)(1.460)(1.880)(0.802)(1.276)
Yield curve spread0.2610.2880.2770.1690.0010.2650.142
(1.824)(2.036)(2.111)(1.340)(1.462)(2.077)(0.172)
Call rate (first difference)0.0020.0020.0040.0020.0000.0010.001
(1.529)(1.428)(2.275)(3.058)(0.748)(1.233)(0.816)
Gensaki rate (first difference)0.1890.2820.2650.2180.2190.3180.156
(1.215)(1.991)(2.044)(1.786)(1.594)(2.564)(1.234)
Ten-year rate (first difference)0.0020.0010.0010.0010.0010.0000.001
(1.376)(0.529)(0.421)(0.831)(0.791)(0.283)(0.670)
Real interest rates
Call rate0.0030.0030.005-0.002-0.0010.000-0.002
(2.329)(2.608)(3.739)-(2.240)-(1.366)(0.402)-(1.844)
Gensaki rate0.0040.0040.0060.137-0.0010.376-0.002
(2.826)(3.089)(4.134)(1.065)-(1.614)(3.166)-(1.750)
Ten-year rate0.0020.0030.003-0.003-0.001-0.002-0.003
(2.069)(2.200)(2.329)-(5.228)-(2.063)-(2.244)-(2.883)
Call rate (first difference)0.0020.0030.005-0.002-0.0010.000-0.002
(1.799)(2.337)(3.435)-(2.195)-(1.526)(0.386)-(1.856)
Gensaki rate (first difference)0.0030.0040.0050.113-0.0010.365-0.002
(2.103)(2.635)(3.610)(0.910)-(1.646)(3.251)-(1.603)
Ten-year rate (first difference)0.0020.0020.003-0.003-0.001-0.002-0.002
(1.846)(2.005)(1.939)-(5.415)-0.979)-(2.417)-(2.599)

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

Table 9.10.Contemporaneous Correlations, 1980-901Coefficient (t-statistic)
Real Growth IndicesInflation Indices
Real GDPTotal domestic demandTotal private demandCPICore CPIGDP deflatorConsumption deflator
Monetary aggregates
M2+CDs (first difference)0.3440.4710.495-0.036-0.091-0.253-0.007
(2.340)(2.642)(2.506)-(0.330)-(1.114)-(2.047)-(0.063)
M2 (first difference)0.3470.4930.498-0.056-0.123-0.304-0.012
(2.613)(3.109)(2.775)-(0.559)-(1.553)-(2.592)-(0.112)
M1 (first difference)0.0640.0350.100-0.094-0.068-0.133-0.055
(1.406)(0.586)(1.657)-(2.530)-(2.740)-(3.636)-(1.347)
Real monetary aggregates
M2+CDs (first difference)0.3350.5380.519-0.332-0.260-0.351-0.313
(3.375)(4.873)(3.950)-(4.766)(5.067)-(4.646)-(3.758)
M2 (first difference)0.3270.5320.519-0.307-0.257-0-298-0.277
(3.485)(5.244)(4.209)-(4.684)-(5.201)-(3.954)-(3.561)
Ml (first difference)0.0590.0640.068-0.127-0.083-0.147-0.094
(1.517)(1.215)(1.135)-(4.288)-(4.072)-(5.150)-(2.661)
Interest rates
Call rate0.0000.0000.0010.0030.0010.0020.002
(0.357)-(0.167)(0.492)(3.467)(1.980)(2.581)(2.599)
Gensaki rate0.1530.0930.0640.1460.3410.0270.766
(0.153)(0.093)(0.064)(0.146)(0.341)(0.027)(0.766)
Ten-year rate0.001-0.001-0.0020.0010.0010.0010.002
(0.543)-(0.738)-(0.947)(1.010)(1.672)(0.692)(1.570)
Yield curve spread0.2060.2890.2220.1230.2070.2490.073
(1.183)(1.793)(1.377)(0.786)(1.175)(1.563)(0.500)
Call rate (first difference)0.0020.0020.0030.0010.0000.0010.001
(1.435)(1.086)(1.921)(1.283)(0.529)(1.362)(0.823)
Gensaki rate (first difference)0.1880.2790.1130.1020.2270.2730.034
(1.034)(1.770)(0.670)(0.754)(1.475)(1.738)(0.211)
Ten-year rate (first difference)0.0020.0000.0020.0010.0010.000-0.001
(1.174)(0.133)(0.928)(0.844)(1.253)(0.219)-(0.517)
Real interest rates
Call rate0.0010.0020.003-0.002-0.0010.000-0.002
(0.841)(1.266)(1.844)-(2.162)-(1.994)(0.417)-(2.006)
Gensaki rate0.0020.0030.005-0.038-0.002-0.001-0.002
(1.637)(1.945)(2.454)-(0.240)-(2.484)-(0.397)-(2.040)
Tea-year rate0.0020.0020.002-0.003-0.002-0.002-0.002
(1.659)(1.598)(1.255)-(4.241)-(2.714)-(2.293)-(2.493)
Call rate (first difference)0.0010.0020.003-0.001-0.0010.001-0.001
(1.324)(1.636)(2.334)-(1.646)-(1.473)(0.881)-(1.609)
Gensaki rate (first difference)0.0020.0030.004-0.033-0.0010.3310.033
(1.644)(1.879)(2.418)-(0.204)-(1.701)(2.356)(0.201)
Ten-year rate (first difference)0.0020.0020.002-0.003-0.001-0.001-0.002
(1.901)(1.747)(1.249)-(4.227)-(2.147)-(1.151)-(2.240)

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

The figures represent the estimated coefficients and t-statistics from the regression of vt on et, where vt is the error on the real growth/inflation equation, and et is the error from the equation explaining the evolution of the financial variables, from the bivariate VARs described in the text.

  • There is a strong positive contemporaneous correlation between real growth indices and the growth of the real monetary aggregates (particularly M2+CDs), which seems invariant to the sample period chosen. The significance of the nominal monetary aggregate is less strong and less stable.

  • There is also a strong positive contemporaneous relationship between the real growth indices and the measures of real interest rates. The significance of the nominal interest rates is less strong and less stable.

  • The positive correlation of both interest rates and the monetary aggregates to real growth could suggest that the dominance of aggregate demand shocks versus policy induced money supply or interest rate shocks.

As regards inflation:

  • There is evidence of a strong negative correlation between inflation (particularly for the core inflation rate) and the monetary aggregates (whether defined in real or nominal terms). This correlation seems relatively stable across different sample periods.

  • There is evidence of a significant positive correlation between the level of the call rate (in nominal terms) and the inflation indices. However, there is evidence of an even stronger negative correlation between real interest rates (whether defined in levels of first differences) and the inflation indices.

Kalman Filter Estimates

The simple, single-equation estimates described in the section rely on several assumptions, including the date at which the shift in the parameters takes place. A more general specification, which allows for a continual evolution of the coefficients, was considered using the Kalman filter approach:

where, again, yt is real GDP and the other right-hand side variables are as defined earlier. The equation was also estimated replacing the log of difference of real GDP with the inflation rate.

Specifically, Figures 9.3 and 9.4 present the results of estimation of a state-space model in which:

Figure 9.3.Kalman Filter Coefficient Estimates: Real GDP Equation1

Source: IMF staff estimates.

1See text for description of the estimates.

Figure 9.4.Kalman Filter Coefficient Estimates: Core CPI Equation1

Source: IMF staff estimates.

1See text for description of the estimates.

The independent variable (yt) was either the rate of growth of real GDP or the rate of change of the core CPI. Xt was a matrix that was assumed to include the lagged (by one quarter) values of the independent variable, the Gensaki rate, the rate of change of the nominal exchange rate index, the rate of growth M2+CDs, and a constant.

The estimates for the real GDP equation were performed over 40 quarters beginning in the 1976:Q2 period, then repeated for each subsequent 40-quarter period. The coefficient estimates (and their standard errors) are those for the final period in each 40-period span. The estimates for the core CPI equation were performed for the 1976:Q2-1986:Q2 period, the coefficient (and standard errors) for 1986:Q1 were saved, then the estimates were rerun for the 1976:Q2-1986:Q2 period, and the coefficients (and standard errors) for 1986:Q2 were saved, and so forth.

The estimates were performed either assuming that the variance/covariance matrix for vt was diagonal (that shocks to one coefficient were uncorrelated with shocks to the other coefficjents), or using a bootstrapped estimate of the variance/covariance matrix. The bootstrapped estimate was calculated by using the estimates of vt from the unbootstrapped equations.

The results for the diagonal variance/covariance matrix are summarized in Figure 9.3. They confirm the instability of the relationship between the nominal financial variables and the indices of aggregate activity and prices. However, unlike the simple regression results above, the Kalman filter estimates suggested an increased role of short-term interest rates as an indicator of the stance of policy since the 1980s. In particular, the coefficient on the Gensaki rate in both real GDP growth and the CPI inflation equations was generally negative and exhibited a downward trend.

The significance of M2+CDs for real GDP growth also appeared to have increased since the 1980s, and the coefficient on M2+CDs growth tended to be higher during the 1990s than in the previous decade. By contrast, the significance of M2+CDs for predicting CPI inflation seemed lower in the latter period. The role of the exchange rate for predicting real GDP growth appeared to be strongest in the 1990s, particularly in 1995, the exchange rate’s peak. By contrast, the exchange rate’s role in predicting the CPI inflation rate seemed to have waned over the sample period. The results of the more general specification, in which estimates of the variance/covariance matrix were “bootstrapped,” are illustrated in Figure 9.4. These estimates were roughly similar to those of the simpler specification. They did suggest, however, a considerably lower level of confidence in the parameter estimates.

The results of the CPI equation suggested a stronger and more consistent role for the monetary aggregate in forecasting the inflation rate. In particular, the coefficient on the growth of M2+CDs was positive and significantly different from zero in both the bootstrapped and unbootstrapped version of the model. Interestingly, in both cases, the coefficient estimates appeared to be on a declining trend, suggesting that the usefulness of the monetary aggregate had waned since the 1970s. The coefficient on the exchange rate was consistently negative, as would be expected, but was generally insignificantly different from zero, particularly in the latter part of the sample period. The short-term interest rate appeared not to provide significant information regarding next-period inflation; the coefficient on the Gensaki rate was often incorrectly signed and usually insignificantly different from zero.

References

    BattenDallasMichael P.BlackwellIn-SuKimSimon E.Nocera and YuzuruOzeki1990The Conduct of Monetary Policy in the Major Industrial Countries: Instruments and Operating Procedures IMF Occasional Paper 70 (Washington: International Monetary Fund).

    BaumgartnerJosefRamanaRamaswamy and GöranZettergren1997“Monetary Policy and Leading Indicators of Inflation in Sweden” IMF Working Paper 97/34 (Washington: International Monetary Fund).

    ChadhaBankim and EswarPrasad1996“Real Exchange Rate Fluctuations and the Business Cycle: Evidence from Japan,”IMF Working Paper 96/132 (Washington: International Monetary Fund).

    ChinnMenzie and Michael P.Dooley1996“Monetary Policy in Japan, Germany and the United States: Does One Size Fit All?” (unpublished; Washington: International Monetary Fund).

    CorkerRobert1990“Wealth, Financial Liberalization, and the Demand for Money in Japan,”Staff PapersInternational Monetary FundVol. 37 (June) pp. 41832.

    Economic Planning Agency1993Economic Survey of Japan: 1992–1993 Chapter 1 and Appendix Notes 1–10 (Tokyo: Economic Planning Agency).

    FriedmanBenjamin M. and Kenneth N.Kuttner1992“Money, Income, Prices, and Interest Rates,”American Economic ReviewInternational Monetary FundVol. 82No. 3 pp. 47292.

    International Monetary Fund1996Japan—Recent Economic Developments IMF Staff Country Report 96/90 (Washington: International Monetary Fund).

    HoshiTakeoAnil K.Kashyap and DavidScharfstein1993“The Choice Between Public and Private Debt: An Analysis of Post-Deregulation Corporate Financing in Japan,”NBER Working Paper 4421 (Cambridge, Massachusetts: National Bureau of Economic Research).

    HuZuliu1993“Yield Curve and Real Activity,”Staff PapersInternational Monetary FundVol. 40 (December) pp. 781806.

    ItoTakatoshi1989“Is the Bank of Japan a Closet Monetarist? Monetary Targeting in Japan, 1978-1988,”NBER Working Paper 2879 (Cambridge, Massachusetts: National Bureau of Economic Research).

    KasaKenneth and HelenPopper1995“Monetary Policy in Japan: A Structural VAR Analysis,”Federal Reserve Bank of San Francisco Working Paper PB 95–12 (December).

    LipworthGabrielle and GuyMeredith1998“A Reexamination of Indicators of Monetary and Financial Conditions,” in Structural Change in Japan: Macroeconomic Impact and Policy Challengesedited byBijanAghevliTamimBayoumi and GuyMeredith (Washington: International Monetary Fund) pp. 15767.

    MadiganBrian F.1994“The Design of U.S. Monetary Policy: Targets, Indicators, and Information Variables,” in Frameworks for Monetary Stability: Policy Issues and Country Experiencesed. by Tomás J.T.Baliño and CarloCottarelli (Washington: International Monetary Fund) pp. 11136.

    McCallumBennett1993“Specification and Analysis of Monetary Policy Rule for Japan,”Bank of Japan Monetary and Economic StudiesVol. 11 (November) pp. 157.

    MorenoRamon and SunBae Kim1993“Money, Interest Rates, and Economic Activity: Stylized Facts for Japan,”Federal Reserve Bank of San Francisco Economic ReviewNo.3 pp. 1224.

    OkinaKunio1993“Market Operations in Japan: Theory and Practice,” in Japanese Monetary Policyed. by Kenneth J.Singleton (Chicago: University of Chicago Press pp. 3162.

    SawamotoKuniho and NobuyukiIchikawa1994“Implementation of Monetary Policy in Japan,” in Frameworks for Monetary Stability: Policy Issues and Country Experiencesed. by Tomás J.T.Baliño and CarloCottarelli (Washington: International Monetary Fund) pp. 81110.

    SchinasiGarry J. and MonicaHargraves1993“‘Boom and Bust’ in Asset Markets in the 1980s: Causes and Consequences,” in Staff Studies for the World Economic Outlook (Washington: International Monetary Fund) pp. 127.

    SingletonKenneth J.ed.1993Japanese Monetary Policy (Chicago: University of Chicago Press).

    SoejimaYutaka1995“A Unit Root Test with Structural Change for Japanese Macroeconomic Variables,”Bank of Japan Monetary and Economic StudiesVol. 13 (July) pp. 5368.

    SoejimaYutaka1996“The Long-Run Relationship between Real GDP, Money Supply, and Price Level: Unit Root and Cointegration Tests with Structural Changes,”Bank of Japan Monetary and Economic StudiesVol. 14(December pp. 2352.

    UedaKazuo1993“A Comparative Perspective on Japanese Monetary Policy: Short-Run Monetary Control and the Transmission Mechanism,” in Japanese Monetary Policyed. by Kenneth J.Singleton (Chicago: University of Chicago Press) pp. 730.

    WescottRobert1996“Assessing the Risks of a Credit Squeeze Among Small- and Medium-Sized Enterprises in Japan,”Japan—Selected Issues IMF Staff Country Report 96/114 (Washington: International Monetary Fund) pp. 6992.

These issues have been addressed extensively in recent IMF policy analysis, including the October 1996 World Economic Outlook.

The experience of the United States, where financial markets are considered to be relatively evolved, is that these relationships may be affected decades after deregulation occurs. As an example, M2 was recently downgraded by the U.S. Federal Reserve as a policy indicator, owing to financial market innovation that has increased the ease with which households could shift between bank deposits and other saving instruments.

A bill to amend the Bank of Japan Law was passed by the Diet in June 1997.

This section borrows heavily from the taxonomy contained in Madigan (1994).

“The Role of Monetary Policy,” speech by Yasuo Matsushita, Governor of the Bank of Japan, at the Research Institute of Japan (November 6, 1996).

Bank of Japan credit was an important source of reserves for banks to meet their monthly reserve requirements. Bank of Japan credit is supplied at the official discount rate, which until 1995 tended to be below the market overnight rate. As Moreno and Kim (1993) note, Japanese banks tended to be more active users of central bank credit than banks in the United States.

See Hoshi and others (1993) for a discussion.

Batten and others (1990) suggest that the shift in focus from bank lending to M2 was the result of a weakening of the link between bank lending to the corporate sector and the ultimate objective of monetary policy.

For a discussion, see Sawamoto and Ichikawa (1994), pp. 100-01. Ito (1989) notes that the Bank of Japan, when it announced in 1975 its intention to focus on M2 growth, cautioned that it would not mechanically adhere to a strict M2 growth target, and his statistical analysis confirms that the Bank of Japan did not follow a strict monetarist rule.

Okina (1993) states that the Bank of Japan’s policy “always begins with controlling interest rates in the short-term money markets.”

Indeed, markets for uncollateralized call transactions, treasury bills, and commercial paper only developed during 1985–87. For a discussion, see Okina (1993).

See Sawamoto and Ichikawa (1994) for a description of deregulation in the money market.

Some observers have criticized the Bank of Japan’s use of such a wide range of instruments in its day-to-day operation of monetary policy, on the grounds that it reduces the transparency of its policy objectives. However, Okina (1993) argues that it has been difficult for the Bank of Japan to adopt a more focused approach given the lack of depth and breadth of the money market.

Indeed, many of the recent empirical studies of the monetary transmission mechanism in Japan (e.g., Chinn and Dooley, 1996; McCallum, 1993; and the Economic Planning Agency, 1993) appear not to have considered the implications of this issue for their results.

In a more recent paper, Soejima (1996) extends these results to argue that the previous evidence supporting the hypothesis that money, income, interest rates, and prices are co-integrated is illusory, and can be overturned if the presence of structural breaks is considered.

For 1975–96, the ADF tests would not accept the hypothesis that the data were stationary in first differences. However, in most cases, either the Phillips-Perron test or the Weighted Symmetric test did indicate first difference stationarity. For a description of the data, see Appendix I.

The test results were disregarded for the interest rate series since they are bounded by zero and are unlikely to be truly nonstationary (i.e., have an infinite variance).

Other studies that have used a similar approach in other countries include Baumgartner, Ramaswamy, and Zettergren (1997) for Sweden, and Lee (1996) for Germany. Moreno and Kim (1993) and Ueda (1993) perform similar tests for Japan.

Only bivariate relationships were considered, as this was thought to bias the results in favor of accepting the hypothesis that a financial variable “caused” the output/inflation variable, and to avoid mistakenly rejecting a possible information variable.

Recent studies that have dealt with this issue explicitly in the case of Japan, using a structural VAR approach include Chadha and Prasad (1996), Chinn and Dooley (1996), and Kasa and Popper (1995).

An exception was the residuals from the real interest rate and real monetary aggregate equations, which were often significantly correlated with the residuals from the price and output equations; see Appendix II for a discussion.

The dummy variables were generally found to be significant. The lag length of the bivariate VARs was determined by minimizing the Schwarz information criterion; the optimal lag lengths ranged between one quarter and three quarters. The Gensaki rate, rather than the CD rate, was used in the full-period regression, owing to data availability.

Hu (1993) also presents evidence on the usefulness of the yield-curve spread as a predictor of output growth in Japan and other major industrial countries.

Tests for 1962-96 (which are not reported since the yen did not float until 1972) showed roughly similar results.

The results did not appear to be sensitive to the inclusion of a measure of the output gap as an explanatory variable.

The BIS’s 8 percent capital-adequacy requirement came into effect in March 1993. The effect of asset prices on banks’ capital positions is reviewed in International Monetary Fund (1996), Chapter VII.

Sawamoto and Ichikawa (1994) describe the difficulties that these balance sheet problems presented for monetary policy. See also Wescott (1996) and International Monetary Fund (1996), Chapter IV.

For example, in the context of an Mundell-Flemming model, the dynamic relationship between interest rates and aggregate output would depend on the source of the shock. Kasa and Popper (1995) compare estimates of VARs estimated for 1975—84 and 1985-94, and conclude that the Bank of Japan’s reaction function exhibited a structural shift away from moral suasion toward market-based instruments.

Note that the estimated coefficients only approximate the correlation coefficients; a more exact approximation would require normalizing the errors by the sample standard deviation.

    Other Resources Citing This Publication