The Japanese deflation: has it had real effects? Could it have been avoided?

Abstract

Has deflation contributed to the long lasting stagnation of the Japanese economy? Could the Bank of Japan have stopped deflation by implementing a more expansionary monetary policy? Tentative answers are probably not to the first question, and probably yes to the second question. It is found that the total cost of deflation over the period 1995–2003 has been close to a 1.1% rate of lost GDP. Yet, on the basis of statistical significance and robustness to specification choices, this evidence is not compelling. On the other hand, the estimated positive linkage between nominal base money growth and inflation is significant and robust, even given current economic conditions. However, in order to be inflationary, monetary policy should have been more expansionary than what actually observed, even since the launch of the quantitative easing in 2001.

Acknowledgements

The author is grateful to A. Beltratti and R. Hetzel for comments on a previous version of the paper, and to an anonymous referee for useful comments. The author is also grateful to ISESAO (Bocconi University, Milan) for financial support.

Notes

 The main explanations for the Japanese slowdown suggested in the literature point to inadequate fiscal and monetary policies to bring the economy out of the liquidity trap, depressed investment due to over-investment during the bubble period of the late 1980s and early 1990s and problems with financial intermediation following the bursting of the bubble, and the reduction in potential output growth determined by a productivity slowdown and demographic effects. See for instance Morana (2004) for additional details.

 Yet, there is not agreement in the literature concerning whether a liquidity trap has affected or is affecting the Japanese economy. See for instance Fujiki et al. (2002), Kimura et al. (2002), Orphanides (2004), Bordo and Filardo (2004) for a sceptical view.

 Over the period 1989:2–1990:4 the Gensaki rate increased of 330 basis points, from 4.3% to 7.6%. As a consequence, base money growth started to decline in 1990, becoming negative in 1991 (− 1.4%) and close to zero in 1992 (0.3%). Money growth (M2 + CD) started to decline in 1991 (2%), becoming negative in 1992 (− 0.5%). Since 1993, despite the steady increase in monetary base growth from 4% to 16% (apart from the fall in 2000), money growth has averaged around a value of about 3%. Over the same time period, as measured by the GDP deflator, inflation started to decrease in 1992, with disinflation turning into deflation in 1995.

 Recently, Morana (2004) has provided empirical evidence in favour of the Hayashi–Prescott hypothesis, while Kamawoto (2004) has found that the reduction in productivity growth may be an artefact due to unaccounted features, such as imperfect competition, varying utilization of capital and labour and cyclical reallocation of inputs across sectors, in the computation of the Solow's residual.

 The argument follows from the equalization of the marginal cost of printing money and the opportunity cost of holding money. See also Uhlig (2004).

 Yet, in the presence of nominal rigidities, theoretical results point to an optimal inflation rate equal to zero, or as large as the real interest rate when the costs from holding money balances are taken into account. However, as shown by recent Asian experience, it is possible that even downward nominal wage inflexibility may disappear as agents get used to deflation, so that real wages and the unemployment rate may not increase during deflationary episodes. Hence, nominal rigidities may be more important in an inflationary environment than in a deflationary one (Bordo and Filardo, 2004).

 The contrasting evidence provided by Engelbrecht and Langley (2001) can be explained by the shorter sample employed by these latter authors (1966–1994).

 See also Bordo and Filardo (2004) for arguments in favour of targeting monetary aggregates during deflation.

 Note that GFLS is equivalent to OLS as μ → ∞.

 The overnight rate is the uncollaterized overnight call rate for the period 1985:3–2004:1. For the period 1980:1–1985:2 the collaterized overnight call rate has been used.

 See also Banerjee and Ruseell (2001). Yet, recent evidence would point to long memory as the cause of persistence of the inflation rate in Japan. In the paper we have assessed the robustness of the results to the selection of the integration order.

 The robustness of the findings has been carefully evaluated. See later in this paper.

 It is important to note that this asymmetry could have not been detected by standard constant parameter VAR analysis, which, on the other hand, points to a significant positive impact of a deflationary shock also in the medium-long term.

 Additional GDP growth would have been close to 3% had price inflation been stabilized at an annual rate of 2%.

 Short term rates were reduced gradually between 1990 and 1995 to levels close to zero, and maintained at fractional levels thereafter. The zero interest rate policy was started in 1999, reversed in August 2000 and reversed again in March 2001, coupled with a quantitative easing aimed at increasing banks’ reserves. However, the latter has led to an expansion in reserves of an amount determined by banks’ demand at the zero interest rate, rather than to an increase beyond such a level (Hetzel, 2003). Since October 2003 the quantitative easing policy has been formalized in a commitment to maintain the zero interest rate policy until deflationary pressures have been dispelled (i.e. until the year on year CPI excluding fresh food rate has achieved a level of 0% or a higher sustainable rate). Recently, the assessment of quantitative easing policy has pointed to the efficacy of the policy in keeping short term rates close to zero, supporting the recovery of economic activity (Bank of Japan, 2004). However, as pointed by Okina and Shiratsuka (2004), as a consequence of low economic growth, monetary policy would have failed to reverse deflationary expectations. Moreover, according to Baba et al. (2004), the reduced worth of lenders and borrowers would also be at the root of the failure of the expansionary monetary policy to affect investment, output and prices.

 Recently Hetzel (2004) has further refined this proposal. According to Hetzel (2004), Bank of Japan should set a target in terms of the price level, and use base money growth as an instrument. The objective would be to relate the demand for excess reserves to the price level, rather than allowing the monetary base to depend on the demand for reserves. Bank of Japan would set an objective for the level of excess reserves, to which banks should adjust, on the basis of the deviation of the price level from the target. Portfolio rebalancing by the public would then be obtained by first exchanging all short term assets in the portfolio of the central bank for long term illiquid assets, and then carry out open market purchases of long term bonds and ETFs replicating the Topix.

 Other non-orthodox policies have been suggested by Meltzer (1999b), McCallum (2000), Svensson (2001), Buiter and Panigirtzoglu (2003), Auerbach and Obstfeld (2003). See also Fujiki et al. (2001) and Svensson (2003).

 Nominal monetary base growth has been about 15% in 2001, 19% in 2002 and 16% in 2003. Over the period 1995–2003 base money growth has been close to 90%.

 It is unlikely that the required policy interventions may be regarded as modest, according to the metric of Leeper and Zha (2003). Hence, the actual policy effects may deviate from the predicted effects due to revisions in agents’ expectations, as also pointed out by Lucas (1976). Yet, the results are still valid and may be considered as referring to a worst case scenario, since the updating in agents’ expectations should lead to a more effective inflationary policy. In fact the revision in expectations is in the direction of a positive or less negative inflation rate.

 The VARFIMA model has been estimated in two steps. In the first step the fractional differencing parameter has been estimated for the variables included in the specification, i.e. the inflation rate, the output growth rate, the change in the overnight rate, the monetary base growth rate, by means of semiparametric estimators (Sun and Phillips, 2003; Beltratti and Morana, 2004). In the second step a VAR(1) model has been estimated using the filtered series by means of GFLS, and the VARFIMA (1, d) parameters have been computed exploiting its VAR representation. Specification tests support the selected model. The results of the semiparametric analysis point to the presence of stationary long memory for all the series included in the specification, of a similar order, i.e. d = 0.38 (0.18). The estimated fractionally differencing parameter for inflation is much larger than the one obtained by Baillie et al. (1996) and Conrad and Karanasos (2005), but similar to the one obtained by Baum et al. (1999), albeit these latter authors also find evidence of nonstationarity and almost I(1) behaviour in some cases. For reasons of space detailed results are not included, which are however available upon request from the author.

 Note that the impulse responses plotted in Fig. 7 are for the price level and the monetary base, rather than for the inflation rate and the monetary base growth rate.

 Similar results are reported in Baba et al. (2004), apart from the effects of the output shock on inflation.

 The full set of results is available upon request from the author.