Forex Seasonality: Is it Real?

I have always wanted to write a post about seasonality, but whenever push came to shove, I couldn’t see the point. Besides, I was never sure whether seasonality falls into the scope of technical analysis or whether it made sense to consider in fundamental terms, and for fear of overstretching, I stayed away. Recently, I read a column by Kathy Lien about forex seasonality. In fact, this article was merely an updated version of a nearly identical article that she contributed earlier to Investopedia, but nonetheless I found it informative, and I was finally inspired to address the topic on the Forex Blog.

Basically, Lien’s analysis consisted of examining 10 years of data for a handful of major currency pairs, and picking out the month(s) for each pair in which performance tended to be most lopsided. (Since forex is zero-sum, it should be the case that over a long enough time horizon, the average fluctuation for every pair should sink to ~0%. For other types of securities/investments, this type of analysis might be less viable). She discovered that the USD has tended to rise against in the Yen in July, but to fall in August. Meanwhile, the Dollar has tended to rise against the Euro in January, and fall against the Canadian Dollar in May. A similar study by DailyFX found that the US Dollar has also tended to rise against the Dollars of Australian, New Zealand, and Canada in the month of July.

These numbers are certainly interesting. But, I want to offer a clarification that the authors, themselves, didn’t bother to make. Namely, when making statistical claims about trends, it’s important to perform statistical (and not just visual) analysis. For example, the fact that the authors based all of their conclusions on only 10 years of data means that the case for statistical significance (a mathematical concept which states that a certain result cannot be a product of pure chance) is not as strong as you would think. Given that major currencies have floated since 1973, there is at least 30 years of good data which can and should have been used in the analysis.

For example, Lien observed that the US Dollar rose 80% of the time against the Yen against in the month of July. Given that the sample size (10 years) is only a fraction of the total data (let’s assume 30 years), we can say with 95% confidence (in accordance with statistical theory) that the actual fluctuation is somewhere between 60% and 100%. If you want to be 99% sure, then the interval expands to 53 to 100. To be fair, most traders would be perfectly happy with 95% confidence, and in this case, that means we can be 95% sure that the Dollar will rise against the Yen at least 60% of the time in the month of July. That’s not great, but still better than a coin-toss. If you bet on this trend every July over the next 10 years, then, you can be 95% sure that you will come out ahead