This post is a little bit different from the tutorials and code centric content that usually appear on this site. In fact, there is no code at all in this post! Instead, I am shall attempt to dive into the world of statistical analysis and take a look at the historical performance of the FTSE 100. I am interested to see if there are any correlation between days of the week, certain days of the month and months of the year.
It is at this point I should point out that I am not a data scientist. Take these findings with a pinch of salt. I am simply curious about the subject matter. If there are any professionals out there, please feel free to let me know whether there are any school boy errors in the methodology. If we are never corrected, we will never learn!
If like me you have spent any time on twitter listening to other traders, you will likely have heard one, if not all of the following pearls of wisdom in your twitter feed:
- “It’s the end of the week. Traders will be looking to take off some risk over the weekend“
- “Buy on Monday and sell on Thursday“
- “It is the start of the month, expect institutional money“
- “Get ready for the Santa rally“
These are interesting ideas and sometimes said by reputable people. However, it would be wise to do our own investigation before following a random strangers advice on the internet (*That includes this blog too!). Therefore, this post aims to add credence to or debunk some of these ideas by looking at the FTSE 100’s historical daily/monthly performance since 1995. Should any of the findings have merit, we may have grounds for setting up an algorithm to exploit it. If they don’t, well we have just saved some time on implementing a useless algorithm.
Let’s take a look.
For those interested, this analysis was performed by downloading historical daily data for the FTSE 100 from Yahoo. The data was then fed in a Jupyter notebook for exploration. A Panda’s dataframe was used to filter the data, make calculations, and output the findings to CSV files. The CSV files were then used to create the final charts you see on this page in good old Excel. (No Matplotlib – Sorry!)
If none of that made sense to you, don’t worry about it. You either came here through google and are only interested in the results or have simply not come across some of the super handy Python modules:
For reference they are:
- Jupyter: http://jupyter.org/
- Pandas: http://pandas.pydata.org/
- Matplotlib, as mentioned was not used. However, it is part of Backtrader and widely used by the science community: https://matplotlib.org/
Days of The Week
The first set of charts looks to see if there is any particular day of the week is more likely to close up or down. Since 1995 there have been 5719 trading sessions on the FTSE. 51.83% of those sessions closed up whilst 48.92% closed down. Hypothetically, if you just bet long every day, you would already have a small edge.
The following table breaks down the historical edges when betting long or short on any particular days of the week.
As we can see from the pictures and table, Fridays do seem to have an edge for people going long in nearly all years under examination. The picture for the rest of the week is a little more mixed with Thursday being the second best performer for longs. Should you subscribe to the “Buy Monday, sell Thursday” mantra, you might be giving up some good gains! Even more dangerous would be going short on the belief that people do not want to hold risk over the weekend. Having seen these results, if you blindly applied that tactic, you would be asking for trouble. Having said that, I don’t want to dismiss the idea entirely. I could see traders being less willing to hold positions over the weekend during times heightened geopolitical tensions. However, the scope of this analysis is not wide enough to answer that question.
Start / End of the Month
The second set of charts attempt to answer the question of whether there is a predictable pattern for the dates around the start and end of the month. I was interested in this question not only because I had heard people talking about “institutional money” but also because of what I think is basic human behavior. Assuming most people receive their salaries towards the month, then it would make sense that a good proportion of the people who buy shares would choose to do so shortly after getting paid.
End of the Month
Start of the Month
The first of the month definitely appears to be a good bet going long. Out of 250 months analyzed, the first of the month closed up a whopping 64.31% of the time! Not bad over such a long period. On the other hand, the end of the month seems to have a short bias although it is not quite as pronounced as the long. It closes down 54.33% of the time.
Months of the year
The final set of charts looks at whether there are any correlations between the month of the year and whether it is more likely to close up or down.
There were a total of 273 months analyzed since 1995. Of those, 58.24% closed up vs 41.76% that closed down. As we can see the “Santa rally” really does have some credence. October and February don’t seem to be bad bets over the long run either.
Average Change in Price Per Month
The final chart shows the average change in the value of the FTSE 100 on a monthly basis.
This chart is particularly interesting for me. Going long in April not only tends to close up more often but when it does, the average gain is over 2x as much as the average loss. Conversely, I would also be cautious going long in September. Even though it appears to close up more frequently, the average loss in that month is a massive -6.17%!
If I am honest, I came into this task not expecting to find much of use. I presumed the results would appear to be quite random. However, after taking a look at some of the charts, it is hard not to see some patterns emerging. The start of the month looks to have a definite long edge. A Santa rally happens much more often than not and Fridays are not a bad bet blindly going long. Whether these are purely random correlations is difficult to say. However, I am now much more interested in seeing whether some of these findings can be used in an algorithm and will certainly be watching these periods more closely in my discretionary trading.
Disclaimer: Nothing in this article should be considered trading advice. Approach the data with caution and draw your own conclusions!