Filter rule profits found in foreign exchange markets in the early days of the current managed float persist in later periods, as shown by statistical tests developed and implemented here. The test is consistent with, but independent of, a wide variety of asset pricing models. The profits found cannot be explained by risk if risk premia are constant over time. Inclusion of the home-foreign interest rate differential in computing profits has little effect on the comparison of filter returns to those of buy-and-hold.
IN THE EARLY YEARS of the generalized managed floating that began in March
1973, filter rule profits in excess of buy-and-hold were found for many countries (Logue, Sweeney and Willett [18], Dooley and Shafer [6], Cornell and Dietrich[5]). It was unclear, however, whether such profits indicated inefficiencies. First, it was not clear that risk was adequately handled in such tests, and often risk was ignored. Second, there was no evidence such profits would be available postsample. And third, there was no statistical test of the significance of these profits.
This paper uses the logic of risk/return tradeoffs to analyze the role of risk in such tests and illustrates the logic in a discussion using the Sharpe-Lintner Capital Asset Pricing Model (CAPM), although a Breeden consumption-based CAPM, a Merton intertemporal CAPM, or an Arbitrage Pricing Model (APM)could easily be used. The paper develops a test of statistical significance appropriate to foreign exchange markets (Section I), analyzing the rate of return on foreign exchange speculation less the foreign-domestic interest rate differential. The test explicitly assumes that risk premia are constant over the sample. The excess rates of return observable in using filters in going from a risk-free dollar asset to a risk-free Deutsche Mark (DM) asset persist into the 1980's (Section II), and this is true even when taking account of transactions costs. It turns out that these results for the $/DM case do not depend very much on the interestrate differential, but primarily on exchange rate behavior. This is useful to know because it is often quite difficult to find matching, interest-rate data of high quality. For a sample of nine other currencies, using only exchange rates, Section III shows that the excess returns made in the first 610 days of the float persist in the next 1,220 days, into the 1980's.
(Dooley and Shafer [7] find similar persistence in experiments that do not use buy-and-hold and have no significance tests.)When filter rule profits have been found in spot exchange markets, it has often been argued that the profits are due to risk. This paper uses the CAPM to analyze rates of return both to buy-and-hold and to filter strategies. It turns out that the * Claremont McKenna College and Claremont Graduate School. Arthur D. Warga, Douglas Joines, and Thomas D. Willett offered helpful comments, and Ning-Ning Koo provided research assistance. 163 164The Journal of Finance CAPM may be used to explain the return to both strategies, but not any exce...
