The impact of news in the dollar/deutschmark exchange rate: Evidence from the 1990 s Stefan Krause December 2004 Abstract In this paper I analyse three specificationsofspotexchangeratemodelsbyusingan alternative approach in defining the news variable. In particular, I employ quarterly data of the U.S. dollar / German Deutschmark exchange rate for the period 1991-1998 in order to determine whether the effect of news announcements on the exchange rate is still present in the decade of the 1990 s. The empirical evidence suggests that news do not seem to provide explanatory power for justifying deviations from either the efficient markets hypothesis or the uncovered interest rate parity. Nevertheless, newspaper announcements and news about inflation do contribute significantly in explaining short run departures from purchasing power parity (PPP) with the expected sign, supporting the view that deviations from PPP will arise from new information available in the market. JEL classification: F31, F41, G15 Keywords: Spot and forward exchange rates, news, unanticipated inflation, efficient market hypothesis, purchasing power parity. Department of Economics, Emory University. I want to thank Nelson Mark and Juan Muñoz for useful discussion. All errors are mine. For comments, please contact me at skrause@emory.edu, 404-727-2944 (phone), or 404-727-4639 (fax).
1 Introduction Since the late 1970 s several authors have tested the hypothesis that announcements of new information have an effect on the spot exchange rate market, as it is the case with other asset prices. Looking at evidence from the 1970 s, Frenkel (1981) points out that new information, rumours and news in general will affect expectations and hence, the behaviour of the spot exchange rate. Furthermore, he claims that these resulting fluctuations in the exchange rate cannot be predicted by the lagged forward rate. In his model he defines news as unexpected changes in the interest rate differential, and finds statistical evidence that news affect the dollar/pound and dollar/frank exchange rates, but not the dollar/mark spot rate. For the period between October 1979 and August 1984, Hardouvelis (1988) models the percentage change in a foreign currency price in a given business day as a function of the unanticipated component of fifteen economic series, finding in the case of the dollar/mark price a significant effect of unexpected changes in the following indicators: M1, bank reserves, surcharge rates, durable goods and retail sales. Hogan, Melvin and Roberts (1991) study yet another news variable: U.S. trade balance announcements. For the decade of the 1980 s, using daily data, they find that the forecast error (based on the survey expectations statistics) is significant in explaining changes in the spot rate for the dollar/mark, dollar/pound and dollar/frank exchange rates, with the expected sign (i.e., an unexpectedly high trade deficit causes the dollar to depreciate). Karfakis and Kim (1995) also look at current account news for the case of the Australian dollar, finding as well that worse than expected announcements result in a depreciation of the currency. Krause (1996) uses a similar approach to analyze the impact of news announcements and changes in foreign currency reserves on the nominal exchange rate in Costa Rica. Finally, Edison (1997) considers expectation errors of the following U.S. indicators: consumer price index (CPI), producer price index (PPI), industrial production (IP), retail 1
sales (RS), unemployment rate (UN) and non-farm enrolment (NF), and finds that the dollar/mark rate responds to news in UN and NF for the entire sample (1980-1995) at or below 10% significance level. The author further separates into good and bad news, which results in a significant effect of RS news to the exchange rate as well. In this paper I analyse three specificationsofspotexchangeratemodels,todetermine whether the effect of news on the exchange rate is still present in the decade of the 1990 s, by using an alternative approach in defining the news variables. Section 2 describes the construction of the news indices and section 3 presents the empirical results, with the dollar/mark exchange rate level and its first difference as the dependent variables. Section 4 summarises the main findings and discusses the limitations of the approach used. 2 Constructing the news indices Hardouvelis (1988) and Hogan, Melvin and Roberts (1991) consider daily information to construct the news indices, whereas Edison (1997) and Cavaglia and Wolff (1996) focus on monthly and quarterly data, respectively. I will follow the latter approach, since I m interested in considering the effect of published newspaper articles on the behaviour of partially uninformed agents. By partially uninformed agents I mean agents that do not receive the information about market developments and changes of indicators immediately after they have been published, but rather react to newspaper articles and headlines about the general state of the economy and, specifically, about the foreign trade balance. In general, there will be good news about the economy when paper articles and editorials comment on the economy blooming, unemployment falling, and a better than average growth in the leading indicators index, whereas there will be bad news about the economy if the contrary happens. Similarly, good news about the trade balance will arise when papers refer to a decrease in the trade deficit and conversely when defining bad news about trade balance. The news indices are constructed by collecting the newspaper abstracts on articles 2
relating to economic indicators in the journal index of the OHIOlink database, which engulfs information published since 1989. The search was limited to the period from 1991:1-1998:12, given that Germany joined the European Monetary Union in January 1999. This search resulted in a total of 2883 headlines of economic news. After eliminating the non-relevant headlines, the number of abstracts was reduced to 1148. The economic news index was constructed in the following way: For every month, a value of 1 was assigned in case there have been overall good news about the economy, as defined above; for practical purposes, a month in which only good news or a ratio at or above 3-to-1 between good news and bad news were published, was assigned a value of 1. Analogously, a month in which either only bad news or the ratio of bad news to good news was equal or greater than 3, was assigned a value of -1. Finally, for months in which there were mixed news (ratios smaller than 3) or no news, the value was set equal to 0. For the current account news index the process was simpler, since during the analysed period there were no mixed or contradictory statements. On months in which there were announcements of a smaller trade and/or current account deficit, the assigned value was 1 and months in which newspapers were writing about bigger trade and/or current account deficits were given a value of -1. Consistently, a value of 0 was assigned for months in which no news were published. Giventhatthealternative specifications were tested using quarterly data, the data in both indices was aggregated to obtain quarterly news dummies; each dummy variable would then have an integer value between 3 and 3, depending on the number of months per quarter where good and/or bad news appeared on the headlines. Finally, I also included a continuous variable of news: the difference between expected and actual inflation. For expected inflation I used the data of the University of Michigan Consumer Surveys of Consumers, Survey Research Center (http://www.isr.umich.edu/src), and considered the discrepancy between the actual and expected annual inflation for the last 3
monthofeachquarter. 3 Estimation Method and Empirical Results In this section I consider three different specifications of spot exchange rate models, and proceed to test whether the three news variables defined above have any impact on the level of the exchange rate or the change thereof. The first model looks at the efficient market hypothesis (EM); the second considers uncovered interest parity (UIP) and the last one focuses on deviations from purchasing power parity (PPP). In all cases I use the quarterly dollar-mark rate and define the exchange rate as the Deutschmark value of the U.S. dollar. 1 Hence, an increase in the exchange rate represents an appreciation of the dollar. 3.1 News and Efficient Markets Following Frenkel (1981), and Baillie and McMahon (1989), the relevance of news in the EM model can be tested using the following equation: s t = a 0 + a 1 f t 1 + a 0 2news + ε 1,t, (1) where s t isthelogofthespotexchangerateattheendofthequarter;f t 1 isthelogvalueof the 3-month forward rate, also at the end of the quarter; news is as vector composed of the variables NC (news about the trade/current account deficit), NE (news about the overall state of the economy), UI (unexpected inflation, as definedinsection2);andε 1,t is the residual. The results of estimating (1) through ordinary least squares (OLS) are presented in Table 1. For this particular specification, the news variables appear with the expected sign (good news about the trade deficit and the state of the economy cause an appreciation of the 1 Quarterly data for the spot and forward exchange rates, consumer prices, industrial production indices, money supply and interest rates were obtained from Datastream (1991:I-1998:IV). 4
dollar; while higher than expected inflation causes a depreciation), but are not significant at a 10% level. Another interesting feature is that the coefficient of the lagged forward rate is significantly less than one and the intercept is different than zero, which differs from the findings described by Frenkel (1981). However, since the objective of this paper is to analyse the role of news, I will omit any further characterisation and discussion of this result. 3.2 News and Uncovered Interest Parity The second model incorporates news in the UIP specification. Hence, the equation to be estimated is given by (2) s t+1 = b 0 + b 1 (i t i t )+b 0 2news + ε 2,t, (2) where i t is the 3-month U.S. Treasury-Bill interest rate; i t is the 3-month German FIBOR rate; and news and ε 2,t are as above defined. Results of the OLS estimation of the model are displayed in Table 2. For this particular model, neither of the news variables is significant at the 10% level and, furthermore, the unexpected inflation regressor appear with the opposite sign. Hence, we cannot conclude that the inclusion of these variables provides an explanation for deviations in UIP. 3.3 News and Purchasing Power Parity For the last model I incorporate the news variables as a means to explaining deviations from PPP. Hence, the specification starts with equation (3): s t = p t p t + χ 0 news, (3) 5
where p t and p t represent the CPI in the U.S. and Germany, respectively, expressed in logarithms. Using the conventional money market equilibrium conditions (Engel and Frankel, 1984) and assuming that both the elasticity of income (φ) and the semi-elasticity of interest rate (λ) arethesameforbothcountries,we have: m t p t = φy t λi t, (4) m t p t = φy t λi t, where m t, m t, y t and y t are the logs of U.S. and German money supply and output for each country, respectively. 2 Finally, assuming covered interest parity (CIP) holds results in the following: ft st =(i t i t ). (5) Combining equation (3) and equation (4) yields: s t = m t m t φ(y t y t )+λ(i t i t )+χ 0 news. (6) Substituting the CIP condition of equation (5) into equation (6): s t = m t m t φ(y t y t )+λ(f t s t )+χ 0 news, which implies s t = ψ(m t m t ) φψ(y t y t )+(1 ψ) f t + ψχ 0 news, (7) 2 For the estimation, I consider for both countries the monetary aggregate M1 as the money supply variable and total industrial production index as a proxy for output. 6
where ψ= 1 1+λ. Assuming that the relationship in equation (7) holds up to a constant term we proceed to estimate the following specification through OLS: s t = c 0 + c 1 (m t m t )+c 2 (y t y t )+c 3 f t + c 4 news + ε 3,t. (8) The results of estimating (8) are presented in Table 3a. For this specification all news variables are significant at the 10% level, with UI significant at the 1% level, and all three coefficients have the expected signs. These findings support the view that deviations from PPP will arise from new information available in the market and these innovations will not be completely reflected in CPI changes. Finally, in order to verify the robustness of the above results, I proceed to compute the White corrected standard errors and also consider a model specification that relaxes the assumption that the elasticity of income and the semi-elasticity of interest rate are the same for both countries. The results are presented in Table 3b. These results allow us to note two important observations. First, we find that our previous results are robust to heteroskedasticity corrected errors and the described changes in the assumptions of the money equilibrium parameters. Second, news about the current account and news about the economy become significant at the 1% and 5% level respectively, which contributes in reinforcing the above described results. These observations lead us to conclude that newspaper announcements regarding the economy and unanticipated inflation explain quite robustly deviations from PPP. 4 Conclusions The main results of this paper, as well as some limitations of the analysis, can be summarised in the following points: 7
The news variables, as defined in section 2, do not seem to provide explanatory power for justifying deviations from either the efficient markets hypothesis or the uncovered interest rate parity. Newspaper announcements and news about inflation, however, do explain significantly short run departures from purchasing power parity with the expected sign, supporting the view that deviations from PPP will arise from new information available in the market. These results are robust to corrected errors and modifications in the assumptions of the money market equilibrium parameters. One major limitation of the study is that, given that the expected inflation index is constructed by considering the median expected price change over the twelve months after the survey, inflationary surprises of either sign tend to be positively correlated on a month-by-month basis, which entails systematic error by the agents. Hence, if we are willing to assume that consumers adjust to these errors in a smaller time frame, the interpretation of the significance of the coefficient has to be careful. 8
5 Appendix: Data Sources Data for the spot and forward Deutschmark/U.S. dollar exchange rates, and consumer prices, industrial production indices, M1 and interest rates for the U.S. and Germany were obtained from Datastream. The information used to compute the news indices was collected from newspaper abstracts on articles relating to economic indicators in the journal index of the OHIOlink database. Expected inflation was computed using the data of the University of Michigan Consumer Surveys of Consumers, Survey Research Center (http://www.isr.umich.edu/src). 9
References [1] Baillie, R. and McMahon, P. (1989). The foreign exchange market: Theory and econometric evidence. Cambridge University Press. [2] Cavaglia, S. and Wolff, C. (1996). "A note on the determinants of unexpected exchange rate movements", Journal of Banking and Finance, 20, pp. 179-188. [3] Edison, H. (1997). "The reaction of exchange rates and interest rates to news releases", International Journal of Financial Economics, 2,pp.87-100. [4] Engel, C. and Frankel, J.A. (1984). "Why interest rates react to money announcements: An explanation from the foreign exchange market", Journal of Monetary Economics, 13, pp. 31-39. [5] Frenkel, J. (1981). "Flexible exchange rates, prices and the role of news ", Journal of Political Economics, 89, pp. 665-705. [6] Hardouvelis, G. (1988). "Economic news, exchange rates and interest rates", Journal of International Money and Finance, 7,pp.23-35. [7] Hogan, K., Melvin, M. and Roberts, D.J. (1991). "Trade balance news and exchange rates: is there a policy signal?", Journal of International Money and Finance, 10, pp. S90-S99. [8] Karfakis, C. and Kim, S-J. (1995). "Exchange rates, interest rates and current account news: some evidence from Australia", Journal of International Money and Finance, 14, pp. 575-595. [9] Krause, S. (1996). "Influencia de las noticias en el comportamiento del tipo de cambio nominal: El caso de Costa Rica (1990-1995)", Instituto de Investigaciones en Ciencias Económicas (IICE), Documento de Trabajo No. 190, San José, Costa Rica. 10
Table 1: News and Efficient Markets Explanatory variable Intercept Lagged forward rate NC NE UI Spot rate (w/o news variables) 0.080 2.475 (0.019) 0.839 12.51 (0.000) Spot rate (with news variables) 0.101 2.753 (0.011) 0.787 9.700 (0.000) 0.014 1.397 (0.174) 0.005 1.436 (0.163) -0.643-0.640 (0.528) Adjusted R 2 Durbin-Watson No. of observations 0.838 1.905 31 0.839 1.901 31 11
Table 2: News and Uncovered Interest Parity Explanatory variable Intercept Interest rate differential NC NE UI Change in the spot rate (w/o news variables) 0.003 0.359 (0.722) -0.058-0.251 (0.844) Change in the spot rate (with news variables) 0.012 1.053 (0.302) -0.152-0.499 (0.622) 0.017 1.147 (0.262) 0.007 1.518 (0.141) 2.161 1.587 (0.528) Adjusted R 2 Durbin-Watson No. of observations -0.033 1.575 31 0.198 1.588 31 12
Table 3a: News and Purchasing Power Parity Explanatory variable Intercept Forward rate M1 differential IP index differential NC NE UI Spot rate (w/o news variables) 0.113 1.590 (0.123) 0.783 8.154 (0.000) 0.036 0.513 (0.612) -0.048-0.562 (0.578) Spot rate (with news variables) 0.115 2.256 (0.033) 0.730 10.68 (0.000) -0.008-0.161 (0.873) 0.035 0.549 (0.588) 0.014 1.883 (0.071) 0.005 1.861 (0.075) -3.938-5.693 (0.000) Adjusted R 2 Durbin Watson No. of observations 0.778 1.518 32 0.898 2.48 32 13
Table 3b: News and Purchasing Power Parity (Specification with White Standard Errors) Explanatory variable Intercept Forward rate M1 differential IP index differential US M1 German M1 US IP index German IP index NC NE UI Spot rate (with news variables, φ=φ*, λ=λ*) 0.115 2.582 0.730 11.83-0.008-0.164 0.035 0.639 0.014 2.636 0.005 2.069-3.938-6.195 Spot rate (with news variables, φ φ*, λ λ*) 1.254 1.117 0.659 12.25-0.049-0.657 0.302 1.440-0.464-1.411-0.120-0.585 0.015 2.783 0.005 2.288-4.039-6.460 Adjusted R 2 Durbin Watson No. of observations 0.898 2.48 32 0.899 2.52 32 14