Exchange Rate Expectations: A Survey of Survey Studies

The empirical literature on survey-based exchange rate expectations is briefly surveyed. The literature in general supports the presence of a nonzero risk premium and rejects the hypothesis of rational expectations. The crucial result is that, whereas short-run expectations tend to move away from some long-run “normal” values, long-run expectations tend to move back toward them. If this behavior of short-run expectations increases the volatility of exchange rate movements, there may be a basis for an official measure to minimize short-run exchange rate movements.

Abstract

The empirical literature on survey-based exchange rate expectations is briefly surveyed. The literature in general supports the presence of a nonzero risk premium and rejects the hypothesis of rational expectations. The crucial result is that, whereas short-run expectations tend to move away from some long-run “normal” values, long-run expectations tend to move back toward them. If this behavior of short-run expectations increases the volatility of exchange rate movements, there may be a basis for an official measure to minimize short-run exchange rate movements.

Although it is firmly established that expectations play a central role in the determination of exchange rates, little is known about the exact nature of those expectations. Of course, the problem with empirically testing hypotheses about exchange rate expectations is that expectations are unobservable. In the past, a popular way to get around this problem was to use either the forward exchange rate or the ex post spot exchange rate as a proxy for the expected exchange rate.

There is an obvious drawback to this approach. First, the use of the forward exchange rate presupposes that there is no risk premium, but the absence of a risk premium itself is a hypothesis of interest.1 Second, the use of the ex post exchange rate imposes rationality of expectations, but the nature of expectations can only be determined empirically. Most empirical tests involving exchange rate expectations are thus joint tests of a hypothesis about the degree of risk aversion (or a more structural model of exchange rate determination) and a hypothesis about the nature of expectations.

In order to avoid the joint nature of conventional hypothesis testing, an increasing number of researchers have recently begun to use survey data in tests involving exchange rate expectations. The use of observable survey expectations allows separate testing of an underlying model of exchange rate determination and a hypothesis about expectations. There is strong professional resistance to the use of nonmarket data, perhaps for good reasons. For one thing, there is no assurance that economic agents have enough incentive to disclose their truthful expectations. For another, even if they did, no precise link seems to exist between average (or individual) expectations and the actual exchange rates that are in fact marginal prices in the foreign exchange market. However, in the absence of better alternatives, an empirical literature based on survey data of exchange rate expectations has been expanding in recent years. This paper presents a brief survey of this growing literature.

The paper is organized as follows. Section I summarizes the features of five major sets of survey data used in the literature. Section II presents a few characteristics of survey exchange rate expectations. Sections IIIVI survey, respectively, major empirical results on forward discounts and risk premia, the rationality of expectations, the mechanism of expectations formation, and the relationship between short-run and long-run expectations. The concluding section presents a summary and a few policy implications.

I. Data Sets

Five sets of survey data have been used in the literature (Table 1). The oldest data set comes from the (roughly) annual surveys of the six-month and twelve-month expectations of over 250 monetary officials and other financial market experts, conducted by American Express Banking Corporation (Amex) of London for the period 1976–85. The exchange rates in the surveys were the U.S. dollar rates of five major currencies, namely, the U.K. pound, deutsche mark, Japanese yen, Swiss franc, and French franc. A problem with the Amex data is that the surveys were irregular and conducted by mail, making the timing of surveys somewhat imprecise. The Amex survey was discontinued in 1985.

Table 1.

Major Surveys of Exchange Rate Expectations

article image

For the period between November 1982 and October 1984, the surveys were biweekly and included only two-week and three-month expectations.

Second, the Economist Financial Report (Economist) has been publishing telephone surveys of the three-month, six-month, and twelvemonth expectations of 14 leading international banks every six weeks since 1981. The exchange rates in the surveys are the same as the Amex data, comprising the U.S. dollar rates of the five major currencies.

Third, a similar data set comes from weekly telephone surveys on the one-week and one-month exchange rate expectations conducted by Money Market Services (MMS) of California. Weekly surveys have been conducted, since 1984 in both New York and London, and each sample includes about 30 professional traders. Between November 1982 and October 1984, the surveys were biweekly and consisted of two-week and three-month expectations. The exchange rates in the MMS data are the U.S. dollar rates of four major currencies, namely, the U.K. pound, deutsche mark, Japanese yen, and Swiss franc.

Fourth, since January 1981, Godwins of London has conducted monthly surveys of 50 leading investment managers. The Godwins data contain only the twelve-month qualitative expectations of the effective and the U.S. dollar exchange rates of the pound. That is to say, the survey asks the respondents only their opinions of the direction of expected future change; that is, whether the exchange rate would go up, go down, or remain the same. Taylor (1989), the only researcher to have used the Godwins data, used a subjective probability method to derive a quantitative series of mean expectations in his study.

Finally, semimonthly telephone surveys have been conducted since May 1985 by the Japan Center for International Finance (JCIF), a private institution affiliated with the Japanese Ministry of Finance. The JCIF data contain the one-month, three-month, and six-month expectations of the U.S. dollar exchange rate of the Japanese yen held by 44 market participants in Tokyo. The surveys have consistently included the same respondents classified into six industry groups, consisting of banks and brokers, securities companies, general trading companies, insurance companies, importers, and exporters; the data can thus be used both as panel data and as industry data.

II. Some Characteristics of Survey Data

Before reviewing the empirical literature on survey-based exchange rate expectations, one should take note of general qualitative characteristics of the survey data. Among many such characteristics, this section will summarize three, under the headings of heterogeneity, underprediction, and “twist,” as a background against which major empirical studies can be reviewed.

Heterogeneity

Although in what follows the convention of treating expectations as if they were homogeneous is adopted, the limitation of this standard practice should be noted at the outset. In reality, any survey data will immediately reveal that expectations have a distribution. If the exchange rate of the Japanese yen against the U.S. dollar is taken as an example, the standard deviations in recent years have ranged roughly between 2 yen and 5 yen per dollar (1 percent and 3.5 percent) for one-month expectations, between 3 yen and 8 yen (1.5 percent and 5.5 percent) for three-month expectations, and between 3 yen and 11 yen (1.5 percent and 7.5 percent) for six-month expectations. As expected, there is a tendency for dispersion (as measured by standard deviations) to increase for longer-term expectations, although the size of increase appears to be far less than would be the case if the variance were to remain constant over expectations horizons (Figure 1). Casual observation also indicates that from the time immediately preceding the Louvre accord in February 1987, the degree of dispersion declined significantly but it began to increase again in the first part of 1989.

Figure 1.
Figure 1.

Standard Deviations of Expected Exchange Rates, May 1985 to December 1989

(yen/dollar)

Citation: IMF Staff Papers 1991, 004; 10.5089/9781451956917.024.A007

Source: Japan Center for International Finance (JCIF).Note: Monthly time series of one-month, three-month, and six-month expected exchange rates of the Japanese yen against the U.S. dollar.

This heterogeneity in expectations may reflect, in addition to the usual distributional factors, systematic individual or group effects. On the basis of disaggregated JCIF data, for example, Wakita (1989) and Ito (1990) found significant industry-specific bias in expectations. According to their findings, exporters had expectations of greater yen depreciation (or smaller yen appreciation) and importers expressed exactly the opposite expectations. This systematic expectational bias in favor of one’s interest may indicate either wishful thinking or strategic behavior (to influence the movement of the exchange rate in a desired direction). Wakita (1989) suggests a possibility that such industry-specific bias may reflect private information, while Ito (1990) argues that, to the extent that individuals are not likely to possess private information, the presence of individual effects must reflect the failure of the rational expectations hypothesis.

Underprediction

Another important characteristic of survey data is a general tendency for the expected future exchange rate of all time horizons to follow closely the current spot exchange rate. This means that expected changes in exchange rates as reported in survey data tend to underpredict consistently the extent of actual exchange rate movements; this is demonstrated by Figure 2, which depicts the actual and expected exchange rates of the U.S. dollar against the Japanese yen during the recent period of sharp dollar depreciation. This is another way of saying that much of actual exchange rate change is unexpected and is consistent with a similar conclusion based on forward exchange rates as a measure of expected exchange rates (Mussa (1979)).

Figure 2.
Figure 2.

Actual and Expected Exchange Rates, May 1985 to April 1986

(yen/dollar)

Citation: IMF Staff Papers 1991, 004; 10.5089/9781451956917.024.A007

Source: Japan Center for International Finance (JCIF).Note: Semimonthly time series of the spot and one-month, three-month, and six-month expected exchange rates of the Japanese yen against the U.S. dollar.

For the earlier period of dollar appreciation (that is, from 1981 through early 1985), survey data indicated a persistent underprediction of the extent of the actual dollar appreciation. In fact, the market participants surveyed in general expected the major currencies to appreciate against the U.S. dollar (Table 2). In contrast, during the period of dollar depreciation (that is, from late 1985 through early 1987), the market participants expected a much more moderate depreciation of the dollar; they even expected a sizable appreciation of the dollar against the pound, when the dollar in the event depreciated in subsequent months. This characteristic of expectations is important to bear in mind when the results of rational expectations tests are interpreted.

Table 2.

Actual and Expected U.S. Dollar Depreciations in Selected Sample Periods

article image
Note: Average annualized logarithmic change in basis points (roughly equal to percentage change); an increase in value means a depreciation of the U.S. dollar against the U.K. pound, deutsche mark, or Japanese yen.

Twist

The final important characteristic of survey data is a tendency for longer-run expectations to reverse the direction of the short-run expectations. That is to say, a depreciation tends to be followed by expectations of further depreciations in the short run, but by expectations of moderate reversals (or appreciations) in the long run. This characteristic has been called in the literature a “twist” in expectations.

This tendency becomes conspicuous during periods of sharp exchange rate movement, as in the recent period of dollar depreciation against the major currencies. For example, when we look at the movements of the expected exchange rates of the U.S. dollar against the Japanese yen during the six-month period between October 1985 and April 1986, we note that the market participants surveyed expected the dollar to continue to depreciate over the period of one month but to appreciate over the period of six months (Figure 3). Some have interpreted these results as reflecting the view of market participants that exchange rates are determined by “momentum” models in the short run but return to historical norms over longer periods (see Section VI for a further discussion).

Figure 3.
Figure 3.

Twists in Exchange Rate Expectations, October 1985 to September 1986

(yen/dollar)

Citation: IMF Staff Papers 1991, 004; 10.5089/9781451956917.024.A007

Source: Japan Center for International Finance (JCIF).Note: Semimonthly time series of the spot and one-month, three-month, and six-month exchange rates of the Japanese yen against the U.S. dollar. The expected exchange rates are dated according to the future periods for which the expectations are formed.

III. Forward Discounts and Risk Premia

The presence or absence of a risk premium has been of considerable interest to economists and policymakers because it has far-reaching implications for the substitutability of assets denominated in different currencies and, hence, for the efficacy of sterilized foreign exchange market intervention. The use of survey data allows the direct measurement of a risk premium from the observation of the forward discount (fd), which can be decomposed into the expected currency depreciation and a risk premium:

fdt+j=tt+jfst=(Etst+jst)+rpt,(1)

where tf is the log of the forward exchange rate set in period t, Et is an expectations operator2 based on the set of information available in period t, st(st+j) is the log of the spot exchange rate in period t(t + j), and rpt is a risk premium; an increase in the exchange rate is defined as a depreciation of the domestic currency (the U.S. dollar).

Table 3 presents the decomposition of the forward discount of the U.S. dollar into expected depreciation (as reported in survey data) and the risk premium in selected sample periods. For the most part, the dollar was expected to depreciate against the deutsche mark and the Japanese yen by an amount greater than the size of the forward discount; in contrast, the dollar was expected to appreciate against the U.K. pound by an amount greater than the size of the forward premium. In general, exchange rate expectations were positively correlated with the forward discount (that is, the currencies that were expected to depreciate were at a forward discount).

Table 3.

Decomposition of the Forward Discount on the U.S. Dollar in Selected Sample Periods

article image
Note: Average annualized logarithmic changes in basis points (close to percentage change).

Although descriptive statistics suggest the presence of a risk premium, it is also important to know if the risk premium is significant (in a statistical sense) and if it is correlated with the forward discount. This question is important because it sheds light on the source of the forward discount bias. The forward rate can fail to be an unbiased predictor of the future exchange rate either because of the failure of rational expectations (to be treated in the next section) or because of a risk premium that is time-varying.

The null hypothesis that the correlation of the risk premium with the forward discount is zero can be tested by running the following regression:

Etst+jst=a1 +a2fdt+j+et,(2)

where e is a random error term. The null hypothesis is that a2 = 1.

According to the results of Froot and Frankel (1989) as selectively summarized in Table 4, the hypothesis of a2 = 1 was rejected for one-month expectations but could not be rejected for expectations of three months or longer (see Sections V and VI for the difference between short-run and long-run expectations). There is thus little evidence that a risk premium is correlated with the forward discount, at least for longer-run expectations. However, the F-statistics rejected the joint hypothesis of a1 = 0 and a2 = 1, suggesting the presence of a nonzero risk premium (this is consistent with the results in Table 3). The failure to reject the hypothesis of a2 = 1 may be interpreted as indicating evidence of the perfect substitutability of assets denominated in different currencies, to the extent that a change in expected exchange rates is fully reflected in a one-to-one change in forward exchange rates, at least for changes in fundamentals observed during the sample period.

Table 4.

Forward Discounts: Summary Results

article image
Source: Froot and Frankel (1989).Note: See equation (2).

All against the U.S. dollar. See Table 1 for the currencies included in each data set.

Standard errors in parentheses; (**) indicates that the hypothesis of a2 = 1 is rejected at the 1 percent level of significance.

F-statistics on the joint hypothesis of a1 = 0 and a2 = 1; (**) indicates that the hypothesis is rejected at the 1 percent level of significance.

IV. Rationality of Expectations

The overwhelming majority of the empirical literature is concerned with the rationality of survey exchange rate expectations. Normally, rationality is defined in terms of two criteria: (1) whether the expected exchange rate is an unbiased predictor of the future spot exchange rate (unbiasedness); and (2) whether the expected exchange rate fully incorporates all available information (orthogonality). The tests of rational expectations reported in the literature also correspond to these two types.

Unbiasedness

Unbiasedness is an important aspect of the rationality of exchange rate expectations. The use of survey data allows direct testing of the hypothesis that the expected spot exchange rate for period t + j (formed in period t) is an unbiased predictor of the future spot rate (in period t+j):

st+j=b1+b2Etst+j+ut,(3)

where the survey expectations Etst+j is free from the presence of a risk premium, and u is a random error term. Tests of the unbiasedness of exchange rate expectations would involve tests of the hypothesis of b1 = 0 and b2 = 1, when equation (3) is estimated, usually in first-difference form.

Dominguez (1986) and Ito (1990) regressed actual depreciation on expected depreciation using MMS and JCIF data, respectively, for different time horizons and for different dollar exchange rates (Table 5). For the earlier period (1983–85), Dominguez almost unanimously rejected the joint hypothesis of b1 = 0 and b2 = 1 for one-week, one-month, and three-month expectations for all currencies.3 The negative estimates of b2 for some exchange rates suggest that the forecasts missed the direction of exchange rate movements. Moreover, the estimate of b2 was below unity in many cases, implying the tendency of forecasters to overpredict the size of future dollar depreciations.

Table 5.

Unbiasedness: Some Regression Estimates

article image
Note: See equation (3). Standard errors are in parentheses; (*) indicates rejection of the unbiasedness hypothesis at the 5 percent level of significance, and (**), rejection at the 1 percent level.

Against the U.S. dollar.

X2-statistic on the joint hypothesis that b1 = 0 and b2 = 1.

Average data.

For the later period (1985–87), however, Ito (1990) could not reject the joint hypothesis except for the six-month expectation. The difference between the two studies may reflect the extraordinary nature of the earlier sample period. As noted earlier, the period studied by Dominguez was one in which the U.S. dollar continued to appreciate on a sustained basis despite expectations to the contrary. Given the extremely low values of R2 in all of these studies (not reported in the table), only a small portion of actual exchange rate changes were predicted in practice. The exact outcome of empirical tests of the unbiasedness hypothesis is thus likely to depend on the sample.

Orthogonality

Orthogonality is another important aspect of the rationality of exchange rate expectations. If expectations are to be efficient (in the sense that they incorporate all available information), their predictable power cannot be improved by inclusion of any variable that is already in the set of information available at the time when the expectations are formed. That is to say, prediction errors must be uncorrelated with any variable in the set of known information. This orthogonality condition can be formally tested by running the following regression:

Etst+jst+j=c1+c2Xt+vt,(4)

where the left-hand-side variable is a prediction error, Xt is a set of information known in period t, and v is a random error term; popular candidate variables for Xt have included forward discounts (or nominal interest rate differentials) and lagged exchange rates. The orthogonality hypothesis is that c1= c2 = 0.

Table 6 summarizes some of the regression results, along with the choice of Xt variables and samples, reported in the literature. While the summary is by no means exhaustive, it gives an indication of the range of results that have been obtained from survey studies. As a general rule, given the large standard errors, t-tests often failed to reject the separate null (orthogonality) hypothesis of c1 = 0 or c2 = 0. However, x2-or F-statistics almost unanimously rejected the joint orthogonality hypothesis of c1 = c2 = 0, particularly for time horizons longer than three months. These results, taken together, seem to suggest that the expected exchange rates as reported in the survey data did not fully incorporate all available information.

Table 6.

Orthogonality: Some Regression Results

article image
article image
Note: See equation (4). Standard errors are in parentheses; (*) indicates rejection of the orthogonality hypothesis at the 5 percent level of significance, and (**), rejection at the 1 percent level.

Against the U.S. dollar.

X2-statistics on the joint hypothesis that c1 = c2 = 0; for Froot and Frankel (1989) only, F-statistics are reported.

All the currencies in the sample are pooled.

V. Mechanisms of Expectations Formation

Regardless of whether or not expectations are rational, it is of interest to investigate how they are formed. Survey data have been used to test three broad types of expectations and their formation mechanisms, according to the classification popularized by Frankel and Froot (1987a): extrapolative, adaptive, and regressive. It should be noted in this exercise that no attempt is being made to determine which of the three expectations mechanisms is correct or even closest to the actual process.

Extrapolative Expectations

The first mechanism is called extrapolative expectations:

Etst+jst=g(ststj)(5a)

or

Etst+j=hst+(1h)stj,(5b)

where h ≡ (1 + g). This mechanism is called extrapolative for the obvious reason that the expected currency movement for the next period is given by the past currency movement for the most recent period, as indicated in equation (5a). Equation (5a), however, can be equivalently expressed as equation (5b), which indicates that an extrapolated expectation of the spot exchange rate for period t + j is given by a weighted average of the current spot exchange rate and the lagged exchange rate for period t – j.

Of crucial interest is the sign of the coefficient g: g < 0 is the case of distributed lag (where past currency movement is followed by an expectation of currency movement in the opposite direction); g = 0 is the case of static expectations (where currency movement is expected to follow a random walk), and g > 0 is the case of bandwagon expectations (where past currency movement is followed by an expectation of currency movement in the same direction). In the case of bandwagon expectations, g > 1 indicates that expectations are explosive.

Table 7 summarizes some of the major regression results on the signs and magnitudes of g reported in the literature. First, in almost all cases the t-statistics rejected the hypothesis of static expectations (that is, g = 0); this means that, despite the fact that empirical exchange rates followed a process that is closely approximated by a random walk (Mussa (1979) and Takagi (1988)), market participants did not expect the expected future exchange rates to follow the same process.

Table 7.

Extrapolative Expectations: Some Regression Results

article image
Note: See equation (5a). Standard errors are in parentheses; (*) indicates that the coefficient is significant at the 5 percent level of significance, and (**), at the 1 percent level.

Against the U.S. dollar.

All the currencies in the sample are pooled.

The right-hand-side variable is always (st –st–2) in biweekly data, regardless of the choice of j.

Second, the sign of g is positive (that is, bandwagon expectations) for the short-run horizons of one or two weeks and one month; the sign of g is negative (distributed-lag expectations) for the long-run horizons of six and twelve months; and the sign of g was mixed for the three-month expectations. This behavioral difference between short-run and long-run expectations is a general characteristic of survey expectations that will show up repeatedly throughout this section. For the magnitude of the estimates, the regression results indicated that the absolute values of g were less than unity in all cases. This means that, for short-run expectations of the bandwagon type, expectations were stabilizing.

Adaptive Expectations

The second mechanism is called adaptive expectations, in which expected currency movement is determined as a fraction of the current prediction error:

Etst+jst=k(Etjstst)(6a)

or

Etst+j=(1k)st+kEtjst.(6b)

As equation (6b), an alternative expression, indicates, the adaptively formed expected exchange rate is given by a weighted average of the actual current exchange rate and the expected current exchange rate.

Adaptive expectations, however, have not been very popular in empirical investigations. Table 8 summarizes some of the empirical results on the sign and magnitude of k that are reported in two studies by Frankel and Froot (1987a, 1987b). First, the signs of k again tend to differ for different time horizons, although the distinction between short-run and long-run expectations is less clear-cut than in the case of extrapolative expectations. For the MSS data containing the expectations over one-week to three-month horizons, the sign of k was unanimously negative, indicating that unanticipated appreciation for the current period (that is, an increase in value of the right-hand-side expression in equation (6a)) leads to continued expected appreciation (that is, a fall in value of the left-hand-side expression).

Table 8.

Adaptive Expectations: Some Regression Results

article image
Note: See equation (6a). Standard errors are in parentheses; (*) indicates that the coefficient is significant at the 5 percent level of significance, and (**), at the 1 percent level.

Against the U.S. dollar.

All the currencies in the sample are pooled.

On the other hand, for the long-run expectations of three to twelve months, the sign of k was generally positive; the negative estimates in Frankel and Froot (1987a) were statistically insignificant. Although these results are not conclusive, they are consistent with an expectations mechanism under which unanticipated appreciation leads to expected depreciation in the longer run. This may reflect the belief of market participants that there is a long-run “normal” level for the exchange rate. In all cases, the absolute values of all the coefficients were less than unity, suggesting that expectations were stabilizing.

Regressive Expectations

The last expectations mechanism discussed in the literature is called regressive expectations, and has the general form:

Etst+jst=q(stst)(7a)

or

Etst+j=(1q)st+qst,(7b)

where st is the long-run equilibrium exchange rate. This mechanism is called regressive for the obvious reason that the actual exchange rate is assumed to regress toward the equilibrium exchange rate (at the speed of adjustment given by q); in this formulation, the expected exchange rate can also be expressed as a weighted average of the current exchange rate and the long-run equilibrium exchange rate.

Obviously, the estimate of q would depend on which model is used to specify the equilibrium exchange rate (st); popular candidates in the literature have been constants, moving averages, and purchasing power parity (PPP) exchange rates. If the sign of q is found to be positive, on the one hand, the expectation is regressive, such that the exchange rate is expected to move in the direction of the specified long-run equilibrium rate; on the other hand, a negative sign for q would indicate an expectations mechanism under which the exchange rate is expected to deviate from the long-run equilibrium rate.

Table 9 summarizes some of the representative regression results reported in the literature. In general, the sign of q was found to be negative for the short-run expectations of one week to one month; the sign of q was positive for the long-run expectations of six and twelve months (except when the coefficient was statistically insignificant); and for the three-month expectations, the signs were ambiguous. The negative sign of q for the short-run expectations suggests that exchange rate expectations can be destabilizing in the short run. Once again, the same striking contrast in behavior is evident between short-run and long-run expectations as in the case of extrapolative and adaptive expectations.

Table 9.

Regressive Expectations: Some Regression Results

article image
Note: See equation (7a). Standard errors are in parentheses; (*) indicates that the coefficient is significant at the 5 percent level of significance, and (**), at the 1 percent level.

Against the U.S. dollar.

All the currencies in the sample are pooled.

VI. Short-Run and Long-Run Expectations

Regardless of which formation mechanism is assumed, short-run (generally shorter than one month) and long-run (generally longer than three months) expectations have been shown to display strikingly different behavior. There is a tendency for short-run expectations to respond to lagged exchange rate movements in the same direction or move away from some long-run “normal” values, while long-run expectations tend to respond to lagged movements in an opposite direction or move toward the long-run normal values. This behavioral difference has generated considerable interest in recent years in terms of how to characterize such behavior (“consistency”) and how to explain it (“chartists and fundamentalists”).

Consistency

As a way to capture the apparent behavioral difference between short-run and long-run expectations, Froot and Ito (1989) recently proposed an analytical concept called consistency, which is a weaker condition of rationality; it is weaker because it does not require that expectations have a certain relationship with the actual exchange rate process. It is important to note, however, that consistency is a model-based concept, such that rejection of the consistency hypothesis is a rejection of the joint hypothesis of consistency and a particular expectations formation process.4

To simplify exposition, assume that the expected one-period depreciation from t to t + 1 is determined by the lagged one-period depreciation from t – 1 to t, as follows:

Etst+1st=w1(stst1).(8)

Equation (8) expresses the formation mechanism of short-run expectations. By updating equation (8), the expected one-period depreciation from t + k – 1 to t + k can be expressed as

Et(st+kst+k1)=w1k(stst1).(9)

This means that the expected fc-period depreciation from t to t + k can be expressed in terms of the one-period lagged depreciation from t – 1 to t, as follows:

Et(st+kst)=Σi=1kEt(st+ist+i1)=Σi=1kw1i(stst1).(10)

Equation (10) is the expected K-period expectation from t to k, as obtained from sequentially updating the expected one-period expectation by k times. Now, assume that the expected k-period depreciation from t to k can also be directly determined by the one-period lagged depreciation from t – 1 to t, as follows:

Et(st+kst)=wk(stst1).(11)

Equation (11) is the formation mechanism of long-run expectations, analogous to equation (8) for short-run expectations. Consistency, defined by Froot and Ito (1989), requires that the K-period expectation obtained iteratively from short-run expectations (equation (10)) equal the same k-period expectation obtained directly (equation (11)). In terms of this simple model, consistency thus imposes the following cross-equation restriction when equations (8) and (11) are estimated:

wk=Σi=1k(w1i).(12)

The restriction (12) indicates that, as long as the absolute value of w1 is less than unity, the sign of wk must be the same as the sign of w1, indicating that the “twist” observed between short-run and long-run expectations in the previous section cannot be consistent. However, such a conclusion is valid only if one accepts an autoregressive process of order one as the correct expectations formation process. For an alternative time-series process, it can be shown that consistent expectations can generate a twist. Tests of consistency are thus conditional upon the hypothesis about the correct expectations formation process.

Froot and Ito (1989) estimated a system of either two or three equations subject to the cross-equation consistency restrictions; they used both first-order and second-order autoregressions to describe the expectations formation process. According to their estimation results (Table 10), consistency could not be rejected for the immediate horizon of one-week to one-month expectations, indicating that these very short-run expectations behave in a similar fashion. In contrast, for the horizon encompassing three, six, and twelve months, one to three months, or one to six months, the cross-equation consistency restrictions were unanimously rejected for the expectations of different horizons at least two months apart. Froot and Ito attributed this finding to the observation that short-run expectations tend to overreact relative to longer-run expectations when the exchange rate changes.

Table 10.

Consistency: Summary Results

article image
Source: Froot and Ito (1989).

Against the U.S. dollar.

Wald statistics on the cross-equation restrictions imposed by consistency; (**) indicates that the restrictions are rejected at the 1 percent level of significance.

Chartists and Fundamentalists

As a way to explain the difference between short-term and long-term expectations, Froot and Frankel (1990) suggested that participants in the foreign exchange market may be using two types of forecasting techniques (see also Frankel and Froot (1986, 1988)). It may be that, for the short-run forecasts, the predominant method is “chartist” or technical analysis; and that, for the longer-run forecasts, the common method is fundamental analysis based on such “fundamental” variables as PPP. They present evidence based on annual Euromoney surveys that the weight of chartists in the market has been increasing in recent years. To the extent that, at least in the short run, the market is dominated by chartists who concentrate on the recent pattern of price movements, their increasing influence may have been a factor in the increased volatility of exchange rates in recent years.

It is, however, too early to give a final verdict on the cause of exchange rate volatility. A recent study by Allen and Taylor (1989), based on a survey of over 200 foreign exchange market practitioners in London, questions the notion that chartism necessarily increases volatility. According to this study, Allen and Taylor find that, while chartism is the most actively used forecasting method over short-time horizons (intraday to one week), it is by no means used exclusively. Both the chartist and fundamentalist methods are often used in a complementary manner, with the latter assuming greater weight at longer horizons. Moreover, there was no evidence that chartist expectations overreacted to changes in the current spot exchange rate and were thus destabilizing.

VII. Conclusions and Some Policy Implications

This paper has presented a brief survey of the empirical literature on survey-based exchange rate expectations. After summarizing the features of survey data sets and the broad descriptive characteristics of survey exchange rate expectations, the paper reviewed the empirical literature under the headings of forward discounts and risk premia, rationality of expectations, mechanisms of expectations formation, and short-run and long-run expectations.

Three broad generalizations emerged from the survey of the empirical literature. First, survey data generally indicated the presence of a nonzero risk premium, which appeared to be stable and uncorrelated with the forward discount. This means that changes in the forward discount fully reflect changes in the expected exchange rate, so that assets denominated in different major currencies can be considered to be perfect substitutes, at least for changes in fundamentals observed during the sample period.

Second, empirical tests were generally unfavorable to the hypothesis that exchange rate expectations are rational in terms of both unbiasedness and orthogonality. Except for certain time periods and horizons, survey expectations were shown to be biased predictors of future exchange rates, and the forecast errors were correlated with some variables that are known to be in the set of information available when the expectations are formed. Given the extraordinary nature of some sample periods, however, the rejection of the rationality hypothesis may be saying more about the peculiarity of actual exchange rate movements than the nature of exchange rate expectations.5 Lewis (1989), for example, suggests that systematic forecast errors can be consistent with the behavior of rational agents who are learning about the new process governing fundamental economic variables.

Third, the most substantive conclusion that emerges from the literature concerns the consistently observed behavioral difference between short-run (generally shorter than one month) and long-run (generally longer than three months) expectations. Short-run expectations tend to respond to lagged exchange rate movements in the same direction and move away from some long-run “normal” values, while long-run expectations tend to respond to lagged movements in an opposite direction and move toward the long-run normal values. This suggests the possibility that the foreign exchange market in the short run reflects an element of “noise trading,” trading that is based on factors other than “fundamentals.”

The conventional wisdom among economists has long held that such noise traders, who base their trading on factors other than market fundamentals, would on average buy high and sell low, and thus would be driven out of the market. De Long and others (1987), however, have recently proposed a model in which such noise traders may survive in the long run, even if they do buy high and sell low on average. In this model, noise traders are rewarded with a higher return for the greater risk they assume; moreover, the greater risk introduced by noise trading would lead rational investors to demand a higher return on risky assets, such that the asset prices can deviate from their fundamental values. Noise trading could also cause a greater volatility of price movements relative to what is warranted by the movements of long-run fundamentals.

If it is indeed the case that a major portion of the short-run volatility of exchange rate movements is attributable to the chartist nature of short-run exchange rate expectations, the literature may provide justification for some type of policy measure to intervene in the foreign exchange market in the short run. Such a measure might be a fixed transactions tax in the foreign exchange market, which would increase the cost of short-run trading relative to long-run trading. A majority of economists, however, would probably remain skeptical of such a policy recommendation until they could be convinced of a firm theoretical link relating average expectations to marginal prices in the foreign exchange market.

REFERENCES

  • Allen, Helen, and Mark P. Taylor, “Charts, Noise and Fundamentals: A Study of the London Foreign Exchange Market,” CEPR Discussion Paper No. 341 (London: Centre for Economic Policy Research, September 1989).

    • Search Google Scholar
    • Export Citation
  • Bank of Japan, Research and Statistics Department, “External Balance Adjustment and Monetary Policy Management under International Private Capital Flows” (Tokyo: Bank of Japan, November 1989).

    • Search Google Scholar
    • Export Citation
  • Boughton, James M., “Exchange Rates and the Term Structure of Interest Rates,” Staff Papers, International Monetary Fund, Vol. 35 (March 1988, pp. 3662).

    • Crossref
    • Search Google Scholar
    • Export Citation
  • De Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann, “The Economic Consequences of Noise Traders,” NBER Working Paper 2395 (Cambridge, Massachusetts: National Bureau of Economic Research, October 1987).

    • Search Google Scholar
    • Export Citation
  • Dominguez, Kathryn M., “Are Foreign Exchange Forecasts Rational? New Evidence from Survey Data,” Economics Letters, Vol. 21, No. 3 (1986), pp. 27781.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Frankel, Jeffrey A., and Kenneth A. Froot, “Understanding the Dollar in the Eighties: The Expectations of Chartists and Fundamentalists,” Economic Record, Supplement (1986), pp. 2438.

    • Search Google Scholar
    • Export Citation
  • Frankel, Jeffrey A., and Kenneth A. Froot, (1987a), “Using Survey Data to Test Propositions Regarding Exchange Rate Expectations,” American Economic Review, Vol. 77 (March), pp. 13353.

    • Search Google Scholar
    • Export Citation
  • Frankel, Jeffrey A., and Kenneth A. Froot, (1987b), “Short-Term and Long-Term Expectations of the Yen/Dollar Exchange Rate: Evidence from Survey Data,” Journal of the Japanese and International Economies, Vol. 3 (September), pp. 24974.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Frankel, Jeffrey A., and Kenneth A. Froot, “Chartists, Fundamentalists and the Demand for Dollars,” Greek Economic Review, Vol. 10 (June 1988), pp. 49102.

    • Search Google Scholar
    • Export Citation
  • Froot, Kenneth A., and Jeffrey A. Frankel, “Forward Discount Bias: Is It An Exchange Risk Premium?” Quarterly Journal of Economics, Vol. 104 (February 1989), pp. 13961.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Froot, Kenneth A., and Jeffrey A. Frankel, “Exchange Rate Forecasting Techniques, Survey Data, and Implications for the Foreign Exchange Market,” IMF Working Paper 90/43 (Washington: International Monetary Fund, 1990).

    • Search Google Scholar
    • Export Citation
  • Froot, Kenneth A., and Takatoshi Ito, “On the Consistency of Short-Run and Long-Run Exchange Rate Expectations,” Journal of International Money and Finance, Vol. 8 (December 1989), pp. 487510.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ito, Takatoshi, “Foreign Exchange Rate Expectations: Micro Survey Data,” American Economic Review, Vol. 80 (June 1990), pp. 43449.

    • Search Google Scholar
    • Export Citation
  • Lewis, Karen K., “Changing Beliefs and Systematic Rational Forecast Errors with Evidence from Foreign Exchange,” American Economic Review, Vol. 79 (September 1989), pp. 62136.

    • Search Google Scholar
    • Export Citation
  • MacDonald, Ronald, and Thomas S. Torrance, “Some Survey-Based Tests of Uncovered Interest Parity,” in Exchange Rates and Open Economy Macroeconomics, ed. by Ronald MacDonald and Mark P. Taylor (Oxford and New York: B. Blackwell, 1989).

    • Search Google Scholar
    • Export Citation
  • Mussa, Michael, “Empirical Regularities in the Behavior of Exchange Rates and Theories of the Foreign Exchange Market,” Carnegie-Rochester Conference Series on Public Policy, Vol. 11 (1979), pp. 957.

    • Search Google Scholar
    • Export Citation
  • Takagi, Shinji, “On the Statistical Properties of Floating Exchange Rates: A Reassessment of Recent Experience and Literature,” Bank of Japan Monetary and Economic Studies, Vol. 6 (May 1988), pp. 6191.

    • Search Google Scholar
    • Export Citation
  • Taylor, Mark P., “Expectations, Risk and Uncertainty in the Foreign Exchange Market: Some Results Based on Survey Data,” Manchester School of Economic and Social Studies, Vol. 57 (June 1989), pp. 14253.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wakita, Shigeru, “Are Survey Forecasts Trusted? American Trade Account Deficit and Yen/Dollar Rate,” Economics Letters, Vol. 29, No. 4 (1989), pp. 33944.

    • Crossref
    • Search Google Scholar
    • Export Citation
1

For example, if there is no risk premium in the exchange rate of any two currencies, assets denominated in one currency are perfect substitutes for those denominated in the other, making sterilized foreign exchange market intervention ineffective.

2

Note that expectations (expressed by the operator E) may or may not be mathematical expectations.

3

Her study also included the Swiss franc and two-week expectations.

4

Boughton (1988) suggests an interesting possibility that short-run and long-run markets are segregated in terms of market participants. According to this interpretation, the expectations of foreign exchange participants in each market can be rational even it they are not consistent with those in the other market.

5

The rejection may also reflect the so-called peso problem which is a finite sample bias attributable to the failure of correct expectations to materialize during the sample period. In the earlier period of dollar appreciation, for example, the consistent bias in expectations might have reflected the expectations of rational agents who correctly perceived the dollar to be “too high.”

Long-Run Money Demand in Large Industrial Countries: Volume 38 No. 1
Author: International Monetary Fund. Research Dept.
  • View in gallery

    Standard Deviations of Expected Exchange Rates, May 1985 to December 1989

    (yen/dollar)

  • View in gallery

    Actual and Expected Exchange Rates, May 1985 to April 1986

    (yen/dollar)

  • View in gallery

    Twists in Exchange Rate Expectations, October 1985 to September 1986

    (yen/dollar)