Cheng Hsiao
Shorenstein Reports on Contemporary East Asia
Volume 2, Number 1
December 1999
The 1999-2000 Shorenstein Seminars on Contemporary East Asia opened the series with a talk by Professor Cheng Hsiao on December 1, 1999, at the Institute of East Asian Studies. Professor Cheng Hsiao's presentation is based on a paper of the same title co-authored with Robert Dekle and Siyuan Wang. The presentation tries to answer whether high interest rates had the desired effect of appreciating the nominal exchange rate in the Asian crisis countries. The authors use Korean high-frequency data during the crisis and its aftermath to examine the relationship between the increase in interest rates and the behavior of exchange rates.
Cheng Hsiao is Professor of Economics at the University of Southern California. He is co-editor of the Journal of Econometrics and serves on the advisory committee of the Institute of Economics of the Academia Sinica in Taiwan. He has served on many other boards and advisory committees. He has authored and edited several books and numerous articles, including Econometric Models, Techniques and Applications, 2nd edition, with M. Intrilegator and R. Bodkin (Prentice-Hall, 1996).
In this paper we tried to look at the relationship between interest rate movements and exchange rate movements. As you all know, the crisis began in June of 1997, when the currencies of Asia's developing economies began to fall rapidly against the United States dollar. This foreign exchange rate and financial crisis in Asia caught many by surprise since these countries were considered to be fundamentally sound, having enjoyed decades of high growth. In 1996, just one year before the crisis, the economic growth rate was 7.1% in South Korea, 8.0% in Indonesia, 8.6% in Malaysia, and 5.5% in Thailand. Nor was inflation, though somewhat higher than in the U.S., by any means out of hand, being 4.9% in South Korea, 7.9% in Indonesia, 3.5% in Malaysia, and 5.9% in Thailand. Government budgets were also more or less balanced, and financial markets were quite buoyed. Thus the sudden drop in investor confidence seems unwarranted. These countries did suffer from some problems, including weak financial systems, excessive foreign borrowing by domestic enterprises, and a general lack of transparency. Yet none of this would seem significant enough to explain the sudden change that occurred in 1997.
Once the crisis hit, these countries began having problems servicing their foreign debt and typically asked the International Monetary Fund for assistance. One of the IMF's stated objectives in giving aid to these countries was to stabilize their currencies could, since depreciation allowed them to sell their goods more cheaply than their foreign competitors, creating an imbalance against their competitors. Moreover, the IMF wanted to make sure that creditors to these nations were repaid. So the typical IMF conditions included reducing the country's growth rate, reducing government spending and raising taxes to create a government surplus, and raising interest rates to stabilize the exchange rate. The IMF also aimed to strengthen financial-sector regulations by increasing transparency and opening markets to foreign participants. Economists generally favored these later objectives, but disagreed about, which some thought might aggravate the problem or perhaps impose a cost too high to justify the hoped-for benefits.
Specifically, we'll look at the IMF's condition that these countries raise their interest rates through tight monetary policy. The argument for such action in the face of depreciation is simply that raising the domestic interest rate increases the rate of return vis-a-vis foreign countries and causes investors to move money into these countries in crisis, thus reversing the capital outflow and stabilizing the exchange rate. However, there is also an argument that raising the interest rate raises the debt service cost of domestic companies; such a situation can cause otherwise stable and well-managed companies might go bankrupt. These bankruptcies would in turn worsen banks' balance sheets and thereby exacerbate, rather than alleviate, the crisis. Thus some, including Jeffrey Sachs of Harvard and J. Stiglitz at the World Bank argue that raising interest rates was actually destabilizing, rather than stabilizing, the exchange rate.
The major mechanisms linking the interest rate and the exchange rate, in both the traditional and revisionist views, can be expressed by the equation relating the current exchange rate and the expected future exchange rate. If you think the market is in equilibrium and there is no uncertainty, then you would expect the difference between the domestic interest rate and the foreign interest rate to be equal to the difference between the future exchange rate and the current exchange rate. For example, if the current and future exchange rates were the same, and the domestic interest rate were higher than the future exchange rate, foreigners could move their money into the domestic market, earn a higher rate of return, and then convert their money back into their national currency. Such movements would be expected to have two effects. First, the flow of money into the country would increase the supply of funds and would thus put downward pressure on the interest rate. Second, the supply of foreign currency would increase, and without a similar increase in the demand for foreign currency the domestic currency would appreciate. Thus in a certain world and competitive economic environment, one expects a direct relationship between the difference in domestic and foreign interest rates and the difference in the current and future exchange rates.
However, in an uncertain world it is less clear what the relationship will be. When uncertainty exists in exchange rate movements investors will demand compensation for this uncertainty in the form of an exchange rate risk premium. So even if the domestic interest rate is higher than the foreign interest rate, foreign investors might not move their money into the domestic market because of uncertainty about what the exchange rate will be in the next period. Moreover, investors might demand a default risk premium if there is a higher probability of default in one country than in another. So you typically find that the difference between domestic and foreign interest rates reflects not just a difference in the current and future spot exchange rates, but also a risk premium.
The standard argument that raising interest rates will stabilize the domestic currency requires the exchange rate and default risk premiums to remain more or less constant. But people like Joseph Stiglitz have been arguing that this assumption is unjustified. If the domestic interest rate rises, then it is possible for the risk premium, to change. Since the rise in domestic interest rates raises the debt service costs of domestic enterprises, it could increase their risk of bankruptcy and thus cause a rise in the default risk premium required by foreign lenders, neutralizing any hoped for pressure for appreciation. Moreover, in this situation, banks will now have large numbers of bad loans and will be forced to cut their lending, thereby causing a credit crunch in the domestic economy. This credit crunch in turn will lead to a fall in aggregate demand, and thus further worsen the domestic economic circumstances. So increasing the domestic interest rate might not serve as a useful instrument in stabilizing the exchange rate.
In this paper we are not presenting a theoretical model to explain what happened, but simply looking at the data for South Korea to see if raising the domestic interest rate had the effect of stabilizing the exchange rate as would be expected the conventional view.
We start with an overall look at the relationship between the exchange rate differential (the future spot rate minus the current spot rate) and the interest rate differential (the interest rate in the United States minus the interest rate in South Korea) between September 1997 and August 1998. There is a significant rise in the Korean and U.S. interest rate differential starting from November, and from mid-December it jumps substantially and stays high until mid-February or March, and then slowly begins to decline. The relationship between the Korean won and the United States dollar (i.e., the won per dollar ratio) is stable until September, then jumps up (i.e., the won becomes cheaper vis-a-vis the dollar) and stays high until February, when it gradually begins coming down. So looking at the broad picture, there might appear to be some justification for the revisionist view in the fact that both the interest rate and the exchange rate differentials seem to have risen and fallen together over this period.
However, if you look more closely at the timing of the change, you see that the exchange rate differential jumps before the interest rate goes up, and it begins to fall before the interest rate comes down. Similarly, you might say that the stabilization of the exchange rate occurs after the rise in the interest rate, so that if you look at the lead-lag relationship between exchange rates and interest rates, the picture might lend support to the traditional point of view. Thus just looking at the graphical relationship gives us no clear idea of the nature of this relationship, nor can it give us any sense of the relative strength of these movements. Moreover, if you look at the relationship between the interest rate differential and the bankruptcy rate in Korea, you find that when the differential is very high it does seem to raise the bankruptcy rate; yet even when the differential comes down, it appears that the bankruptcy rate continues to fluctuate at a constant level. So again, the strength and closeness of this of this relationship is unclear from a mere graphical analysis. Thus the goal of our statistical analysis is to more precisely understand these relationships.
Looking at the relationship between the interest rate differential, the exchange rate differential, and the default premium (measured by the difference between the futures market rate and the current spot rate), we see that the rise in the default premium seems to match more closely with movements in the exchange rate differential than with movements in the interest rate, and when the interest rate begins to come down, the risk premium does not actually fall along with it. We might also define the risk premium in terms of the difference between the price of the United States Treasury bond and the price of Korean-issued, dollar-denominated long-term bonds, where the difference in price can be thought to reflect the higher risk of default on the part of the Korean government. Again, except for some correspondence during the high interest rate period, the interest rate moves up and down, but the default premium remains more or less constant. So changes in the interest rates appear to have little effect on people's expectation of bankruptcy and default. Still, graphical representations are subject to various interpretations.
Doing a statistical analysis of these relationships requires that one posit a theoretical model of the relationship between these variables. Various theoretical models have been proposed in the literature, and this paper also includes one. However, it is important to keep in mind that these are simple theoretical models. They are designed to capture the most significant elements of a relationship while ignoring elements that might be considered less relevant. Ideally, they investigate the relevance of different variables until arriving at a satisfactory degree of realism. Nevertheless, such simplifications could undermine the validity of a statistical analysis, and although we use our simple model to show how the relationship between exchange rates and interest rates might be interpreted differently with different parameter values, we feel that the model is still too simple.
So in the empirical analysis we did a time-series analysis of the relationship between exchange rates, interest rates, and risk premium variables which may correspond to different theoretical model with the imposition of proper theoretical constraints. We just looked at the relationship between the variables to see if we could make any conclusions from it. What the statistician does is try to find relevant summary information that describes the essential relationships among the data, such as co-movement between the variables at different lags. Alternatively, one might posit a structural relationship iwht some disturbance term that presumably captures all that is not explicitly accounted for in the model. In the case as well, one looks at the data a-theoretically to gauge its relationships and see if they can fit into the framework of the proposed model. If the proposed model can be estimated to fit the empirical data well, then one simply assumes that the simplified model represents the true data-generating process. Thus the time series approach goes from a theoretical analysis to a specified theoretical model. If one had an infinite number of observations, then this approach would lead to unambiguous conclusions. Unfortunately, with only a finite number of observations this approach cannot be said to give the "true" model, and so it leads to debates among statisticians as to which model is closest to the true one.
Doing various time-series approaches, we found that there is some justification for the IMF's position that raising the interest rate has an effect on stabilizing the exchange rate. Also, we did not find any strong statistical relationship between the interest rate and the bankruptcy rate. However, even with this fairly weak relationship between the interest rate and the exchange rate, the effect is so small that it is questionable whether it is a useful policy instrument for stopping exchange rate depreciation. We found that stopping depreciation by raising the interest rate requires a very large rise in the interest rate differential; and even though the relationship between the interest rate and bankruptcy rate is unclear, it is hard to imagine such a large rise in interest rates not having detrimental effects on the long-run growth of the Korean economy. Raising the cost of doing business for companies by such a large amount might not increase their likelihood of bankruptcy, but it almost certainly will lead to less investment and thus less growth of the economy.
Overall, we might not find evidence for the revisionist view that raising interest rates could actually worsen currency depreciation, but neither do we find evidence that such action will significantly stem such depreciation.
Since the end of the Bretton Woods regime, economists have tried to see what kinds of structural models might predict the movement of exchange rates. The amazing thing is how miserably these models fail. If you're going to predict a future exchange rate, it turns out that the best prediction to give is the current exchange rate. If I put in different interest rates or relative money supply or relative GNP, in a flexible exchange system, the variables do not predict the exchange rate well at all. This might seem obvious given uncertainty, yet even if you put in ex post the realized values of GNP or money supply, you still don't do a good job explaining exchange rate fluctuation. Thus even with perfect foresight, these variables just don't explain exchange rates well.
One of the usual stories, along the lines of the Lucas Critique, says that these models are treating expectations too crudely and that the parameters of those models actually depend on people's expectations. Thus when people see big shifts in exchange rates and big shifts in interest rates, the models' parameters shift because people change their assessment of what the policy rules are. So there is a whole line of research that tries to take account of how expectations change, developing sophisticated models to deal with this phenomenon. Yet as far as I know, they too do a pretty poor job of predicting exchange rates.
In the case of the Asian economies affected by the currency crisis in 1997, there is some good news and some bad news as far as analysis goes. The good news is that in most cases, at least through 1996, we are not dealing with a flexible exchange rate system, but rather a pegged rate, or "dirty float" system, where the exchange rate is determined by the policy rules of the exchange rate regime rather than the forces of the market. Thus in analyzing these cases, and specifying a model, we feel it is important to take these regimes into account, so that large changes in policy rules and large shifts in exchange rates can logically be seen to affect the expectations of participants in exchange rate markets.
In looking at the data on exchange rate and interest rate differentials in this paper, you find that sometimes the exchange rate is leading the interest rate and sometimes the reverse. This paper takes a co-integrated model of the long run to explain short-run fluctuations in the variables, and when you have expectations playing a role, it is not clear to me the correct frequency over which to associate these variables. Should I be looking at exchange rate and interest rate co-movements week by week or at some other frequency? Certainly the co-integration, which says the relationship between these variable should reach some kind of steady state, will take care of some of this frequency issue, but not all of it.
I would like to comment on the relationship between bankruptcy rates and the interest rate differential explored in this paper. In the United States, we find models trying to estimate the default premium as a function of the underlying structure of the firm, so that, for example, as its assets go down in value ceteris paribus its interest rate differential will go up. In South Korea, if the economy is weak, the sequence goes from a poor economy to poor prospects for companies to a higher default premium. In Professor Hsiao's charts this doesn't show up as an increase in bankruptcies until later. In the United States you usually find that something fundamental happens to a company, and then four or six months later the rating agencies get around to changing their ratings, and then some time after that, perhaps nine months after the initial event, you begin to see an increase in bankruptcies. Expectations of bankruptcy will get immediately reflected in default premiums, well ahead of even the rating agencies' revision of ratings. So I wonder if in Korea one could get more detailed data on corporate spreads, and using that data, if I might not find more of a coincidence between expectations of bankruptcy and movements in these default premiums.
What I understand of what happened in Thailand and Malaysia is that I had a lot of firms borrowing from overseas and borrowing short so that they had to keep rolling over their debt. If a firm is borrowing short term and investing long term, its risk is completely unhedged. Apparently a lot of firms were leaving this risk unhedged, and as soon as exchange rates started depreciating, their borrowing costs went up while the return on their long-term investments was fixed. Thus we have exchange rates becoming exogenous to the interest rate differential rather than vice versa. In this case, a rise in interest rates (being caused by a depreciating currency) would not necessarily increase the costs to a business if it had hedged its exchange rate risk.
So there are two points regarding the relationship between interest rate differentials and bankruptcy. First, if interest rate differentials are being driven by a weak economy, then you should see them affect bankruptcy rates, but with a relatively long lag. Second, if interest rate differentials are driven by exchange rate depreciation, then you might see them lead to increased bankruptcy rates if firms are not hedging against exchange rate risk, in which case bankruptcy is a result of firms having taken bets and lost, rather then being a result of depreciation per se. My understanding of the situation is that the great lack of transparency on the part of firms allowed them to undertake such high-risk behavior. In that case, the source of the problem should be looked for in this lack of transparency rather than simply in the fluctuation of exchange rates.
Summarized by Jason Eis, Ph.D. candidate in the Economics Department at the University of California, Berkeley.
