With the looming US elections, it is tempting to ponder how the markets have reacted during previous election years. Numerous analyses were also done with the South African elections. Common question: Which sectors always perform well during elections and what is the trade if a certain party wins? There are many pitfalls and biases to these types of analyses based on seasonality. In this article, we look at the typical cases of seasonality and highlight some of the pitfalls and why these types of analyses should be done with caution.

Seasonality in equity markets is a phenomenon where stocks or sectors exhibit predictable returns or patterns at certain times of the year. For example, the January Effect states that stocks rise in January because of end-of-year portfolio adjustments and tax strategies. The January Effect, which may have been US-specific, can be generalised as a month-of-the-year effect where returns in a month are outsized compared to other months. Similarly, day-of-the-week and day-of-the-month effects have been researched previously.

Darrat et al. examine seasonal anomalies in the JSE daily returns from 1973 to 2012. They found no January or December effects in the South African market but instead found that returns on Mondays and Tuesdays are significantly lower than on Wednesdays. They also found a beginning-of-the-month effect where returns on the second and third trading days are more pronounced than on other days. Worthington found seasonality in the Australian stock market, over the period from 1958 to 2005, with returns on Tuesdays, returns in December and returns on the second day of the month being more pronounced. More importantly, both papers claim that these seasonal effects have declined or disappeared, with Darrat et al. suggesting that the South African market may have filtered out these anomalies post the 2008 global financial crisis and become more efficient. Worthington mentions that these effects in the Australian market have become less important since the market crash of 1987.

Taking a step back, it is important to understand why these seasonal effects occur and why they may have declined.

Even though there are statistical tests done in both Worthington and Darrat et al., it is difficult to determine why these effects may exist in the first place. In general, these seasonal effects may not be statistically significant across different periods and environments, and they could lead to false signals. The fact that both papers mention that these effects have declined or disappeared also hints at the lack of significance or sustainability of these kinds of signals. Doing more of these types of analyses opens the door to data mining, which means we are bound to think that these signals may be true purely based on the data. Two of the quantitative defences against data mining are to test the financial rationale behind effects and to test these effects across markets to see if there is consistency. Seasonal effects are difficult to test on both these fronts. It is difficult to think of a financial rationale for why markets may consistently perform better on certain days or months compared to other months. Also, we can’t necessarily test these factors across markets because some of these may be country-specific like the January effect in the US.

Another consideration is that usually, people look at the mean or median returns to analyse these effects. So, the claim would be that on average returns for certain days or months are higher than others. But instead of looking at mean or median returns, if we looked at the distribution of returns on those days or months, the effect may go away. If this is the case, the chances of it being a false signal and the effect being spurious are higher.

The following graphs illustrate this point very well. The first two plots are of the mean and median returns from the JSE-ALSI over different days and split into different periods. Purely looking at the means and medians, we can be (falsely) convinced of a Monday effect where returns on Mondays are higher, but also a Friday effect where returns on Fridays are comparatively lower. But as we look at the full distribution with the outliers, our confidence in the existence of these effects must be lowered. Taking the full distribution into account helps us clarify our thinking about these so-called effects.

The evolution of markets and the increase in their efficiency can be contributing factors for these effects to be traded away. This could be the main reason why it is suggested that there was a decline in those factors after market crashes. With the advancement in technology, equity markets have become more efficient in incorporating new information. Globalisation and the increase in diversified global investing strategies mean that local or regional seasonal factors have been diluted even further. Another significant technological improvement has been the rise of algorithmic and high-frequency trading. Accelerated market reactions have reduced the chances of benefitting from seasonal anomalies in general, but more specifically so on longer time frames like days or months.