Financial Innovation and the Stability of Money Demand in Korea

The existence of a stable demand function for M2 in Korea is important for the conduct of monetary policy and financial analysis. However, despite much research into the subject, results thus far yield no conclusion as to whether M2 demand is in fact stable. Previous studies, however, have failed to adequately control for financial innovation. When adding a proxy for such innovation to the cointegrating vector, we find that M2 demand is in fact stable, thus reconciling previous conflicting findings. INTRODUCTION The demand for money in Korea has received attention from a number of authors in recent years. These studies have yielded important insights into the longrun stability of demand for aggregates such as M1 and M2 (see Bahmanin-Oskooee and Rhee, (1994), Lee and Chung, (1995), Lee and Hwang, (1998), Chun (1998)). The aggregate M1 is a fairly narrow definition of money that includes currency and checking account balances. On the other hand, M2 is a broader measure that includes M1 plus savings and time deposits. The purpose of this study is to examine the stability of demand for M2. This category of money was long the intermediate target of the Bank of Korea in setting policy. While it no longer serves such a formal role, there is always controversy over whether central banks should adopt such formal targets (Fatas, Mihov and Rose (2004) find a formal target of any type reduces inflation). Moreover, M2 is still a very significant indicator of liquidity for policymakers and agents in the economy. Ascertaining whether a stable long run demand for M2 exists is therefore important, and has properly attracted the attention of a number of authors. The mixed results thus far on cointegration and the plausibility of parameter estimates imply that further investigation is warranted. In addition, another puzzling aspect of M2 has been the secular downward trend in velocity (see Kim (1992) and Chun (1994)). Velocity is the average number of times a given unit of currency turns over, usually within a year. This presents a bit of a mystery, given that velocity is usually expected to rise over time as the payments system evolves. This paper will also examine the impact of financial liberalization and innovation for obtaining a stable demand function. Financial innovation allows agents to economize, over time, on cash holdings. Tobin (1965) emphasized the importance of innovation to money demand. Other authors such as Baba, Hendry and Starr (1992), Bordo and Jonung (1987) and Lieberman, (1977) have used various Southwestern Economic Review 52 techniques to model the process. Failure to do so may cause false inference in estimation. Accordingly, we will here employ a linear time trend within the cointegratng vector. This specification yields results that are stable and plausible. The other puzzle, falling velocity, has been attributed to the “monetization” of the Korean economy. Several authors ( Yoo (1994), Chun (1998) ) have found that the coefficient on the scale variable in the money demand function exceeds one, implying that real money balances have an income elasticity of greater than unity. Agents are thus willing to hold greater money balances for a given set of income and opportunity cost variables. This effect is contrary to that from financial liberalization, which allows agents to economize on cash holdings. However, that the two effects exist will be demonstrated. To do so requires the modeling of innovation so that innovation and monetization are not conflated, which may cause omitted variable bias as well as a failure to find cointegration. This paper proceeds as follows. The previous results on Korean money demand are surveyed. Then the importance of innovation to money demand is explored. Finally, a long run demand function for M2 is estimated, and results indicate stability and plausible estimates once a linear trend captures the process of liberalization. The coefficient on output is greater than one, implying monetization and explaining falling velocity, while the negative trend implies that there has been a movement in the other direction, towards economizing cash balances, due to financial liberalization. PREVIOUS RESULTS For any nation, stability (or lack thereof) in monetary aggregates has significant policy implications. The decision on what instruments to use as intermediate and operating targets depends crucially on the nature of money demand. Previous studies have yielded important insights into this subject for Korea. These studies examine whether cointegration exists for real money balances and interest rates and income. This analysis begins with a set of variables which are nonstationary. That is, they have no well-defined finite mean or variance. As such, standard regression analysis on such variables would lead to spurious results. However, there could be a linear combination of nonstationary variables that is stationary. For instance, spot currency and forward currency prices are individually nonstationary, but a linear combination is stationary since the two never prices never move far from each other for long. Thus to find stable money demand a researcher must find cointegration between real money balances and other determinants such as interest rates and output. Focusing only on M1, Arrau, et al. (1995), found mixed results when testing for cointegration using the Engle-Granger method. Bahmani-Oskooee and Rhee (1994, referred to as B-R hereafter) found results indicating that M1, and not M2, were cointegrated. Since the Bank of Korea was formally targeting M2 at the time, this suggested that the authorities switch targets as a policy implication. Later studies, such as Yoo (1994), Lee and Chung (1995), and Chun (1998) found cointegration for M2 and in the case of the latter two, none for M1. Lee and Whang (1998) found cointegration for M3, but no meaningful long-term relationships for M2 or M1. These studies have used a variety of techniques and regressors. In B-R, as in Arrau, et al., the Engle-Granger method of detecting a cointegrating vector is used. The others rely on the Johansen test. All rely on some form of income and Financial Innovation and the Stability of Money Demand in Korea 53 opportunity cost variable. B-R and Chung and Lee add the real exchange rate as a determinant of money demand (and do not find stability without it for either aggregate). Chun, in addition to scale and interest rate measures on the right hand side, used the real wage as an indicator of permanent income. These papers indicate that a cointegrating vector likely does exist for M2, given that most studies using the Johansen procedure have found one. Another finding of some of these papers is that real money demand responds more than proportionally to a change in output. While Lee and Chung (1995) find an income elasticity of only 0.602 for M2, Yoo (1994) finds a corresponding figure of 1.13. Moreover, Chun (1998) finds that the effect of transactions on money demand is on the order of 1.3. These estimates are quite plausible when compared to other results worldwide. Goldfeld (1987, p. 137) points out that some studies of money demand in the United States found income elasticities of nearly two. This “excess sensitivity” of M2 holdings to a change in output is presented as evidence that the Korean economy is becoming “monetized”, or that money is playing a larger role in transactions generally. This effect can explain the secular decline in M2 velocity noted by a number of authors (see Kim (1992), Chun (1998), and Lee and Hwang (1998)). In most nations, velocity is expected to increase over time as payments systems evolve and cash management improves. However, the increased willingness to hold M2 as income increases would tend to move velocity in the opposite direction, despite increasing financial sophistication in Korea. MOTIVATION AND METHOD In modeling long run money demand, accounting must be made for the effects of financial liberalization and innovation. Over time, as restrictions on holding interest-bearing assets are loosened and agents learn to economize on holding cash, less money should be demanded for a given set of scale and opportunity cost variables. This phenomenon has been determined as important in a number of empirical studies for many nations. In the U.S., for example, Baba, Hendry and Starr (1992) use a weighted learning curve to proxy adaptation to new financial instruments as a determinant of M1 demand. Arrau, et al (1995) use a random walk term in the cointegrating regression to capture the effects of innovation. Bordo and Jonung (1987) employ the currency-money ratio to measure the effect of financial deepening on money demand. Finally, other observers, such as Lieberman, et al. (1977) include a linear time trend in the demand function to account for the increasing ability to manage cash balances over time. Financial liberalization has been ongoing in Korea since the 1970s. Failure to account for its effect on M2 holdings can impart an omitted variable bias on the estimates of a money demand equation. There have been many cases in the literature on money demand for both developed and developing nations where traditional equations “break down” in that they forecast more cash than is actually held. Moreover, it is often the case, for instance, that in many studies cointegration fails to obtain, or that parameter estimates are of the wrong sign or insignificant. It follows that inclusion of a proper measure of financial innovation can lead to sensible estimates of stable, long run money demand functions. This is an important issue for Korea as there is as yet no consensus on a stable M2 demand function for the country. The idea can be summarized in the following equation: Southwestern Economic Review 54 (M/P)t = β 1yt + β 2it + β 3fit + ε t (1) Here, M is nominal M2, P is the producer price index, y is real output, i is the nominal interest rate, and fi is financial innovation. The key question at hand is finding a proper proxy for such financial innovation in Korea. Since Korea liberalized in a fairly gradual fashion over a long period (see Kim, (1992), and Park (1996) for an accounting of the liberalization process in the late 1970s and 1980s) a time trend in the long run demand function should serve as the appropriate approximation of innovation. Arrau, et al. (1995) employed a random walk term to find monetary stability in Korea for M1, but this technique did not produce stability in this case (although it did for other nations whose payments systems were doubtless far less advanced than Korea’s). It is likely appropriate in nations where there has been greater financial repression and little in the way of liberalization. A weighted learning curve for various financial instruments, as was employed in Baba, Hendry and Starr (1992) would be difficult to model here. The aggregate in question for that paper was M1, and there was only one really relevant change that the learning curve was employed to detect (adjustment to the NOW and Super NOW accounts recently legalized). There were a large number of laws changed with regard to interest rates in Korea, on the other hand, as well as many other changes in the financial sector, and modeling all relevant changes would lead to high collinearity among the regressors. Thus, we will employ a linear trend to capture the effects of gradual innovation. This trend will be part of the cointegrating vector itself, rather than outside of the long-run money demand function, in the VAR, as has been done in previous studies (see Lee and Chung, (1995) as well as Lee and Hwang, (1998)). This is important as the effect of liberalization should enter as a variable directly into the long run money demand function. The hypothesized sign of the trend coefficient is negative. On the other hand, if the Korean economy has become monetized, there is an increasing tendency to hold cash balances, which is contrary to the effect of innovation. This is an unobserved process, and there is no way to explicitly model it based on first principles. A proper measure must, however, capture the main effects of growing sophistication in managing cash holdings. The effect of greater monetization and decreasing velocity should show up, as it has in previous studies, on the income parameter. The impact here is to lower velocity. Again, it is initially puzzling, in the face of financial innovation, to observe falling velocity. An increasing ability to mange money holdings would be expected to raise velocity, as generally occurs. However, the very liberalization which allows agents to minimize cash balances also permits greater interest paid on many categories of money. Thus it is not entirely surprising that velocity has fallen in Korea over the past two decades. A similar phenomenon was observed in the United States, during 1982-1983, when velocity experienced a sharp drop after years of steadily rising. Making this turn of events all the stranger was the fact that those years experienced relatively high nominal interest rates. The factor that seemed to lower velocity was the ability of checking and other deposits to earn interest, which they could not do in earlier years, thus encouraging a longer holding period for cash. This same situation is likely at work in Korea, with liberalization having these contrary influences on money demand. The following model will capture these effects. Financial Innovation and the Stability of Money Demand in Korea 55 DATA AND ESTIMATION The data on M2 was compiled from the International Financial Statistics database and covers the period 1976:4 through 1998:3 (all data is quarterly). The monetary aggregate is deflated by the producer price index. The scale variable is real GDP. Other scale variables have proven useful for some countries, but Park (1998) has indicated that real GDP is the main source of transactions demand in Korea. For the opportunity cost determinant, care must be taken since some studies have included deposit rates that are more properly own-yields and should be expected to have positive signs (see Lee and Chung’s (1995) comment on B-R, p.104 ). Thus, the rate on government housing bonds is used. Lee and Chung (p. 104) discuss the importance of seasonal effects in the Korean money market. Accordingly, all of the variables are first seasonally adjusted before any analysis to avoid any biases that seasonality might impart (M2 is first divided by the price index, as seasonally adjusting only nominal quantities and then deflating can impart biases on the estimates. See Lee and Chung, p. 106, note 7). Next, the Augmented Dickey-Fuller test is applied to all of the variables and the results are displayed in Table 1This test is designed to detect the presence of unit roots. Each series, denoted here as zt, is modeled as follows: Δ zt = α +γ zt-1 + ε t (2) If we cannot reject the null hypothesis that γ = 0, we conclude that the given series has a unit root and is nonstationary. As noted, the null hypothesis of a unit root cannot be rejected at any standard significance level in any case.