Periods when the day of the week anomaly worked


We achieved about 10-14% CAGR over 5 years. That is quite remarkable, but we are set back by a couple of signs. For example, the flimsy D_stat(%) performance. Usually the directional accuracy we achieved is only about 51.5%. (see previous posts) That is a not a good sign.
Let’s see a chart from a backtest of the ensemble ANN. The ensemble is made of 20 ANNs. The lookback period is 200 days. The short term lookback is 20 days. (more about it in a following post). nEpoch = 4; The TR% (Total Return) is this:

What can we conclude? To be candid, I am not too happy about it.
The chart shows the period from about 2004-03 to 2010-03 (6 years).
The day of the week anomaly worked from 2004 to 2005. It stopped working in 2006 to 2007. It worked again in 2008, but there was a problem in 2009, while it also worked after 2009 mid-year.
Ok, we should accept that there are 2 years outperformance and 2 years underperformance. Maybe we cannot avoid that. The problem is however the 2008 financial earthquake. What we can see on this chart is that most of the gain comes from 2008. If we exclude that period, the strategy is in loss. Maybe the anomaly strategy would have worked in 2008 anyway, but the huge volatility of the market (remember the VIX records) multiplied the gain. Without that record volatility, we wouldn’t have been given such a great return in 2008. And do you expect that those exceptional 2008 market environment comes back? We hope not. If not, then should we expect to have 2008-like profit with this strategy in the future? Probably we shouldn’t.
This 2008 phenomenon explains perfectly why we had only 51.5% directional accuracy, while at the same time we had a quite remarkable 10% CAGR. Most of the gains come from the high volatility of the ‘lucky’ 2008 periods and not from the fact that we could forecast the direction of the market accurately.
We had a sad realization today, but we learned something useful. Relying on the final measurements, like Total Return, average CAGR is not enough. Even other statistics, like Maximum Drawdown wouldn’t yell any warning about the pitfalls of the strategy. We have to consider not only the final outcome of these statistics (like TR% as Total Return), but how that statistics evolved over time. We have to plot the evolution and study it and always question what can go wrong in the future…


No Responses Yet to “Periods when the day of the week anomaly worked”

  1. Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: