Does the transition to and from Daylight Saving Time (DST) have a (significant) effect on the stock market?
In a recent blog post on The UK Stock Market Almanac, the author found that the average return of the FTSE100 index for the days following the start of British Summer Time (BST) was -0.07% during the interval 1985-2013. This was lower than both the average for all days during this interval (+0.03%) and for all Mondays during this interval (-0.01%). Is this difference significant?
An article by M. J. Kamstra, L. A. Kramer, and M. D. Levi  concluded that:
the transitions to and from DST have significant effects on the market, a factor of 2 to 5 times larger than the normal weekend effect;
the change in autumn (reverting to “normal” time) has a greater effect than the change in spring (transition to “summer time”). </ul>
In a reply to this article, J. M. Pinegar  pointed out that the significant autumn effect was due to two large outliers caused by international stock market crises. The original authors responded  by showing that, in fact, the entire distribution of returns was shifted to the left. This was the start of a game of paper ping-pong with numerous authors publishing papers which either supported or refuted the Daylight Saving effect.
It’s probably not possible to base a winning strategy on this effect, but it’s worth taking a closer look at the data.
Grabbing Data on Daylight Saving Time
The first thing that we need is the set of dates for the DST transitions. These are available from the National Physical Laboratory web site and I am going to use the XML library to scape the data.
These dates are the Sundays on which the transitions took place. We are interested in the effect on the following day, so we advance them by one day.
Grabbing Data on Market Indices
We will take a look at the effect in the British and American stock market, using the FTSE100 and S&P500 indices as indicators.
We now have two data structures, one for each index. To make the analysis easier, we will consolidate these into a single structure.
The missing values for FTSE100 are because our data for this index only begin in 1984. The data are not currently in a “tidy” format, but that is easily remedied.
Let’s take a global look at the distribution of the returns.
That’s not terribly illuminating. The distributions both look more or less symmetric around 0%. There are fewer counts for the FTSE100 because the data does not go as far back as the S&P500. The scale on the x-axis hints that there are some large outliers, but they are rare and don’t even show up on the histograms.
Now that’s a lot more interesting. We can see that the level of variability changes appreciably from year to year. 1987 and 2008 are particularly volatile years. And the massive negative return in 1987 is the Black Monday event of 19 October 1987.
Okay, let’s focus our attention on the returns for the DST transitions. First we’ll merge the index and DST date data. I am sure that there is a more elegant way of doing it, but this gets the job done.
We are going to make comparisons of the returns on the DST transition days to the average returns for the corresponding year. So we gather the summary parameters for each of the indices, grouped by year.
Next we merge these back into the DST data and form a new standardised variable for the returns.
Finally we are ready to make a plot.
For each index there are two panels, the top one indicates the distribution of standardised returns for the beginning of DST and the lower one for the end of DST. It’s interesting to note that there is a sharp peak at zero for both indices at the beginning of DST. This peak is absent at the end of DST.
The density curves all indicate that the distributions are negatively skewed. The skewness is mostly due to a few large negative outliers. We can find out when these outliers occurred.
So, for both indices the large outliers at the end of DST occurred on 1987-10-26 and 1997-10-27. These dates agree with the outlier dates identified by Pinegar .
The skewed distributions only result in an appreciable shift in the median (indicated by the solid vertical line in the plots) for the S&P500 index at the end of DST. In the other three cases the median is very close to zero. Do the standardised returns differ significantly from zero? We can have a look at the Wilcoxon and t-tests.
The resulting p-values agree with the qualitative analysis: there is only a significant difference (at the 5% level) for S&P500 at the end of DST (agreeing with ).
Spring 2014 DST Transition
As it happens, we have just had a Spring DST transition. Below are the data for the S&P500 and FTSE100 indices for the last two trading days, retrieved from Yahoo! Finance.
Let’s have a look at the returns.
That’s interesting: the S&P500 index experienced a positive return, while the return on the FTSE100 was negative. We know from the analysis above that the return associated with the DST transition can go either way. We also know that the significant effect on the S&P500 is observed for the Autumn rather than Spring transition, so the results from the weekend are not surprising.
M. J. Kamstra, L. A. Kramer, and M. D. Levi, “Losing Sleep at the Market: The Daylight Saving Anomaly,” Am. Econ. Rev., vol. 90, no. 4, pp. 1005–1011, 2000.
J. M. Pinegar, “Losing Sleep at the Market: Comment,” Am. Econ. Rev., vol. 92, no. 4, pp. 1251–1256, 2002.
M. J. Kamstra, L. A. Kramer, and M. D. Levi, “Losing Sleep at the Market: The Daylight Saving Anomaly: Reply,” Am. Econ. Rev., vol. 92, no. 4, pp. 1257–1263, 2002. </ol>