Financial innovation and Divisia money in Taiwan: comparative evidence from neural network and vector error-correction forecasting models

Contemporary Economic Policy, April, 2004 by Jane M. Binner, Alicia M. Gazely, Shu-Heng Chen, Bin-Tzong Chie

I. INTRODUCTION

A standard result of most textbook macroeconomic models that include money and prices is that changes in the money supply lead eventually to proportional changes in the price level, or, alternatively, long-run rates of money growth are linked to inflation. Friedman and Schwartz (1982) present a simple analysis of the correlation between U.S. money and prices over a span of more than 100 years, and Hallman et al. (1991) provide evidence of a long-run link between M2 and the price level using the P-star model based on the long-run quantity theory of money. It appears that the long-run causal chain is just as Friedman said it should be--inflation is a monetary phenomenon.

In the United States, the favored monetary aggregate among monetarists, particularly Milton Friedman, during the early to mid-1970s was simple sum M2. Barnett (1997) paints a very clear picture of the monetarist stance in the United States in the early 1980s in his description of the "broken road." Forecasts showed that the rise in growth of M2 from under 10% to over 30% between late 1982 and early 1983 was bound to result in renewed stagflation, that is, recession accompanied by high interest rates and rising inflation. Friedman's very visible forecast failure, according to Barnett (1997), delivered a very "serious blow to 'monetarism' and to advocates of stable simple sum money demand equations."

Central banks around the world became convinced of the importance of money as a policy control variable and confident in the use of monetary aggregates as intermediate monetary targets just at a time when everything started to go embarrassingly wrong. During the mid- to late 1970s, evidence of the deterioration in the formerly stable demand for money function was beginning to emerge, making the monetarist regin a short one. It was becoming apparent throughout the developed economies in the mid-1980s that increased competition within the banking sector and the computer revolution in the financial world was beginning to have substantial effects on the relative user-costs of bank liabilities and the ever increasing array of substitutes for them. It is now well established that monetary targeting failed in the major macroeconomies because the chosen target aggregates did not remain stably related to other key macroeconomic variables, such as nominal income. Some countries (such as the United States and Germany) moved from narrow to broader money targeting in the mid-1980s before officially abandoning targeting altogether in the late 1980s (e.g., United States and Canada). The Bundesbank kept its monetary goal until the formation of the European Central Bank, although Svensson (2000) claims that the Bundesbank has been an inflation targeter in deeds and a monetary targeter in words only. The consensus of opinion at the end of the 1980s was that it was not possible to reestablish the former apparently stable demand for money functions, even though broad monetary aggregates had been redefined and extended to include higher-interest-bearing building society deposits (see Hall et al., 1989).

Recent research into the construction of monetary aggregates (see the Barnett 2nd Serletis collected volume for the United States) attributes the breakdown in demand for money functions during the 1980s to the use of conventional official simple sum aggregates. Simply summing the constituent component assets to form the aggregate creates flawed index numbers because aggregating any set of commodities with equal weights implies that each good is a perfect substitute for every other good in the group. The simple sum aggregation method will lead to the mismeasurement of the monetary services provided, particularly during periods of significant financial development, when interest rate yields on the various components of broad money are changing over time. The use of equal weights for the user costs of the constituent component assets is wholly inappropriate during periods of high financial innovation because the introduction of new instruments and the technological progress that occurred in making transactions almost certainly have diverse effects on the productivity and liquidity of monetary assets.

Many attempts have been made to improve the measurement of money. The most well-known attempt is that of Friedman and Schwartz (1970). They suggested applying some form of weighting of the components in the aggregate depending on their relative "moneyness." The pioneering work of Barnett (1978, 1980) has provided a consistent method to perform this weighting. Economic aggregation theory provides methods for choosing which assets to include in a monetary aggregate and how to construct aggregator functions, whereas index number theory provides parameter- and estimation-free methods to perform the aggregation. One index that has particularly good properties for the purpose of constructing monetary quantity indices is the Divisia index, derived from the class of superlative index numbers discussed by Diewert (1976). Much attention has also been devoted to the viability of alternative weighting schemes, such as weighting by bid-ask spreads, turnover rates, price variations, and denomination size. The preference for a Divisia monetary index was founded on neoclassical microeconomic theory, approximation theory, and revealed preference theory. When applying these theories to the construction of monetary aggregates, it becomes apparent that the components included should be weighted depending on the monetary services they provide. It can be shown that traditional simple sum aggregation is only justified when all asset components are perfect substitutes (Barnett, 1984).