¹ A description of the exchange rate data as well as the rest of the data used in this paper may be found in the Appendix.

² This was the month when the Canadian government announced in its budget message that it would be government policy to move the Canadian exchange rate to a substantial discount on the US dollar.

³ For a comparison with the post-1918 experience, see my paper, “Speculative Prices as Random Walks: An Analysis of Ten Time Series of Flexible Exchange Rates,” Southern Economic Journal, XXXIII (04 1967), 468–78.Google Scholar

⁴ Jorgenson, Dale W., “Minimum Variance, Linear, Unbiased Seasonal Adjustment of Economic Time Series,” Journal of the American Statistical Association, LIX (09 1964), 681–724.CrossRefGoogle Scholar Professor Jorgenson's model in its multiplicative form assumes that each observation may be represented as the product of trend-cycle, seasonal, and irregular components. Estimates of the seasonal factors are obtained by a generalized least-squares regression applied to the logarithms of the data. The estimated seasonal factors for the first three quarters are the regression coefficients for dummy variables representing the quarters, while the estimated factor for the fourth quarter is the constant term in the regression. The log factors are constrained to add to zero. I am indebted to Professor Jorgenson for a copy of the computer program for the IBM 7090.

⁵ Note that the exchange rate used is the Canadian/US exchange rate, and so higher arithmetic values for this rate reflect depreciation.

⁶ However, the F-ratios for testing the hypothesis that, in the regression used to estimate the seasonal factors, all the regression coefficients are zero are 32.9 and 7.7 for receipts and payments, respectively, and are significant at the .1 per cent and 1 per cent levels respectively. The F-ratio for the exchange rate seasonal factors regression is only 0.18, which is in the lower tail of the F-distribution and is significant at the 2.5 per cent level. The estimated seasonal factors for the exchange rate are, therefore, much closer to zero than one would expect under the null hypothesis of no seasonals.

⁷ The government capital flows examined below are those of the central government only. (See item 3 in the Appendix.)

⁸ It is clear that ratio-to-moving average techniques cannot be used when some observations are negative. The Jorgenson technique could have been used, but the computer program available to me required strictly positive observations due to the assumption of a multiplicative model.

⁹ However, the seasonal factors in Table I suggest that the current account deficits in the first and second quarters should be very nearly equal on average, which they are not according to Table II.

¹⁰ For a more complete discussion of the random walk hypothesis, see Cootner, Paul, ed., The Random Character of Stock Market Prices (Cambridge, Mass., 1964).Google Scholar

¹¹ While most hypotheses of irrational speculative behavior predict either persistence or reversals in prices, it would also be possible for speculators to merely add to the variance of price changes while leaving the serial dependence nearly zero. Since the variance of changes in the Canadian exchange rate was so small, this type of speculative behavior, if it existed, could hardly be considered a problem during the period under study and so will not be discussed further.

¹² Changes in the natural logarithms closely approximate percentage changes for changes as small as those in the Canadian data. Therefore, the convenient phrase “percentage changes” will be used instead of “changes in the natural logarithms” in the remainder of the paper.

¹³ The constant term in the regression is zero to five decimal places.

¹⁴ Alexander, Sidney S., “Price Movements in Speculative Markets: Trends or Random Walks,” Industrial Management Review, II (05 1961), 8.Google Scholar This paper is reprinted in Cootner, , ed., The Random Character of Stock Market Prices, 199–228.Google Scholar

¹⁵ The beginning move was determined by the data. The computer program examined each observation in turn, comparing it with the previously determined peak and trough. If the observation was higher than the previous peak, it defined a new peak, and similarly with a trough. If a new peak was determined, it was then compared with the trough and, if the new peak was x per cent above the trough, a buy signal was defined. Similarly, if a new trough was x per cent below the peak, the first signal would be a sell signal. After the first signal, say a buy signal, the computer program would examine each observation in order to define a new peak or determine whether the observation were x per cent or more below the peak in which case a sell signal would be given. After a sell signal, the program would search for a new trough and a new buy signal given by an exchange rate x per cent or more above the trough. And so the process continued until the end of the series was reached. The program assumed that the hypothetical speculator closed out a short position and went long at the actual market exchange rate on the day after the buy signal, and closed out a long position and went short at the actual market exchange rate on the day following a sell signal.

¹⁶ The average per cent return was rounded up to .02 for presentation in the table, thus explaining the discrepancy between .54 and .634 = 31.7 × .02.

¹⁷ That is, if C_t is the short-term capital flow in quarter t, and d_t is the percentage change in the quarterly average exchange rate between quarter t and quarter t−1, the correlations in Table IV are between C_t and d_t , and C_t and d _t−1.

¹⁸ This is not to say that changes in macro-variables are the only determinants of fluctuations in flexible exchange rate.