The Currency Model

Currencies move in trends. Short-term trends can last a few days. They unfold within trends that last a few weeks, which, in turn, occur within trends of a few months. These medium-term trends unfold into major trends that may last several years. This pattern of trends within trends has existed in currencies since they began to trade freely in the 1970s.

All trends vary in duration and amplitude. It is therefore impossible to accurately forecast a currency’s price at a time in the future. However, Richard M. Levich, Professor of Finance and International Business at New York University, Stern School of Business, has found that " Forecasting currency is not hard when one realizes that accuracy is not essential for most investors. What is important is getting the direction of the forecast right." (Can Currency Movements be Forecasted, AIMR Conference Proceedings, 1998)

A G. Bisset & Company’s currency model was developed in 1983 to make directional forecasts by exploiting the fact that price-changes in currencies are normally distributed.

When price-changes are normally distributed, the flipping of a coin will have a 50% probability of being correct. Bisset’s model has a history of making correct directional forecasts of about 60% in the major currencies. Prof. Richard M. Levich found, "When the percentage-correct track record is significantly greater than 50%, there is evidence of forecasting expertise." (Can Currency Movements be Forecasted, AIMR Conference Proceedings, 1998)

Currency price-changes are normally distributed whether they are measured daily, weekly, monthly, or quarterly. Since each distribution is bell-shaped, half of all price-changes will be negative while the other half will be positive. However, when the correctness of directional forecasts is around 60% the bell-shaped distribution of losses and gains is moved to the right creating more gains than losses as shown in the chart.

The statistical serial correlation in currency prices is very weak. That is why the probability of a price rising or falling, from one day to another, from one week to another, and from one month to another, is 50%. However, even though the serial correlation is weak, currencies move in trends and they will continue to move in trends as long as price-changes are normally distributed. An example will illustrate why trends exist and why they will continue to unfold in the future.

In 2001-2002, the average daily change in the dollar against the euro was –0.02% with a standard deviation of 0.66. As a result, 68% of the daily price-changes fell between –0.68% and +0.64%. In the same two years, the average weekly price-change was –0.09% with a standard deviation of 1.33 that created a range of –1.42% to +1.24% for 68% of all weekly price-changes. Since that range was twice as wide as the daily range, the larger weekly price-changes could only occur because there were series of daily price-changes that created short-term trends.

The average monthly price-change in 2001-2002 was –0.51% with a standard deviation of 2.45. It produced a range of –2.96% to +1.94% for 68% of all monthly price-changes. Since this range was wider than the weekly range, the larger monthly price-changes could only occur because there were series of daily and weekly price-changes that created those monthly price-changes. Thus, even though price-changes are statistically independent of each other from day to day, from week to week, and from month to month, price-trends irrefutably exist. And, they must continue to exist as long as currencies are traded freely and the distribution of their price-changes is normally distributed.

A currency’s price is determined by the demand for one currency over another and is driven by fundamental economic and political factors in the medium- and long-term. However, while the net demand may be positive or negative over one month or longer, the balance between buyers and sellers fluctuates randomly from day to day and from week to week and results in the random price-changes that are normally distributed.

A moving average of daily or weekly prices will smooth the random price-changes and it reveals the underlying flow of money in and out of a currency. When a currency is in demand, its moving average will rise as prices increase. When a currency is out of favor, its moving average will decline as prices fall.

When a moving average is superimposed on a price-chart, it will appear as if prices fluctuate around the moving average. Because prices "rise above" and "fall below" the moving average, they will be at or near their largest positive deviation from the average when an upward price-trend has been sustained and is about to reverse. Likewise, a price will be at or near its largest negative deviation from the average when a price has declined and is about to turn up. Bisset’s model exploits this fact to make directional forecasts.

Because monthly and quarterly price-changes are normally distributed, it follows that a price cannot deviate too much from its moving average. And, because price-changes are normally distributed, the probability of a trend-reversal will increase as prices move away from the moving average.

Bisset’s currency model exploits the fact that currencies move in trends and that the trends reverse direction when a price has attained a large deviation from the moving average. To identify the turning points, Bisset’s model measures the price-momentum with a proprietary algorithm designed to identify trends of two to four months in duration.

When a downward trend ends and changes into an upward trend, a currency’s price momentum becomes less negative, turns up, and begins to rise. As the upward trend unfolds, momentum increases, becomes positive, and then reaches a high level at which it will turn down when a new downward trend starts. These low and high momentum levels coincide with the times at which prices have large deviations from their moving average. Thus, when momentum is low or high, it indicates that the probability is high that a trend-reversal will occur.

Since each trend’s duration and amplitude cannot be known in advance, the model’s momentum measurements are coupled with proprietary decision rules that result in conditional recommendations to buy, sell, or maintain a position. These recommendations are combined with price limits that must be penetrated to cause the model to give a signal to buy or sell.

The price limits acts as a filter that ensures that a price-move is sufficiently large to be more than a random weekly fluctuation within a two- to four-month trend. To further increase the probability that a weekly price decline or rise, which occurs when momentum is high or low, is the start of a trend-reversal, the price limit must be penetrated twice. A double limit penetration will trigger a signal to buy or sell.

A limit penetration occurs when a currency trades at or below the limit at noon in New York on any business day and then again trades at or below the limit the next business day, at 9:00 AM, in New York. As a result, the model can give a signal to buy or sell on any business day.

Once a signal to buy or sell has been given by the model, Bisset’s currency team will implement that signal in the currency market acting as an agent for its currency overlay and currency alpha program clients. The currency model’s focus on two- to four-month trends results in an average of two to three hedges per currency per year.

Top of Page | Contact Us for More Information