With markets finally erm reverting this week, it's a good time to update on the success of the most basic of mean reversion strategies – namely buying after a down day and selling after an up day.

'Mean Reversion' is a nebulous theme, but the core idea is to sell when the market is too hot and buy when its over sold.

Mean Reversion (MR) started to become successful following the tech bubble collapsed and went into over drive during the credit crunch. Since then, things have become a little more pedestrian. This true for both the simple upday/ downday method and more sophisticated methods such as those using the RS2 or DVB. In geneneral the (former) inexortable uptrend didn't do these pull back strategies any favours.

Here's the latest scores on the doors up until Monday 21st.

**Long after a down day on SPY (No commission or slippage). **

Since March 2008: 58% return/ Ave day 0.18%/ 2.22% Standard Deviation.

Last 12 months: 13.3% return/ Ave day 0.13%/ 1.33% Standard Deviation.

Last 5 months: 6.68% return/ Ave day 0.22%/ 0.81% Standard Deviation.

The chart below shows this short only strategy since March 2008.

**Short after an up day on SPY (No commission or slippage). **

Since March 2008: 37% return/ Ave day 0.09%/ 1.59% Standard Deviation.

Last 12 months: -7.2% return/ Ave day -0.05%/ 0.88% Standard Deviation.

Last 5 months: -8.85% return/ Ave day -0.13%/ 0.66% Standard Deviation.

The chart below shows this short only strategy since March 2008.

**SPY Buy and hold. **

Since March 2008: 21% return/ Ave day 0.03%/ 1.90% Standard Deviation.

Last 12 months: 20.5% return/ Ave day 0.08%/ 1.08% Standard Deviation.

Last 5 months: 14.78% return/ Ave day 0.15%/ 0.68% Standard Deviation.

Short selling has not been a fruitful strategy in the recent up trend. You would have had to very carefully pick your spots to make any sort of return in the last 6-12 months.

**A different approach**

One thing that struck me about the recent uptrend was the number of times the bulls were able to drag the index up off the lows. Could this be a better way of looking at activity rather than simple close to close comparisons?

I tested a simple calculation as follows:

(Close-low) / (High-low)

In other words – What percentage of the day's range is accounted for by the difference between the close and the low? The further the day closes away from the low, the closer this percentage will be to 100%. The measure doesn't look at how today's close compares to yesterday's it's all about the abilty of the buyers to drive intraday action higher.

I doubt this is an original calculation, but I still found it of interest.

I used a score of 55% as being the dividing line. Below 55% is negative intraday action and above 55% is positive intraday action. The average day since 2008 is 55%, arguably 50% should have been the cut off.

Let's have a look at a MR strategy using the following rules:

**Buy on a close below 55%:**

Since March 2008: 76% return/ Ave day 0.22%/ 2.17% Standard Deviation.

Last 12 months: 25.36% return/ Ave day 0.26%/ 0.37% Standard Deviation.

Last 5 months: 11% return/ Ave day 0.37%/ 0.74% Standard Deviation.

I won't bore you with another chart, but the returns from going short above 55% are similar but better to the returns from the simple short MR strategy (It has lost less money recently)

This different way of looking at MR improves on the simple upday/downday measure and still leaves room for further application. It's not a massively dramatic improvement on simple MR, but it is an improvement nonetheless

**Next…**

Can this calculation be used as an environment filter for other MR type strategies? It seems so when looking at the average scores over different periods. In other words if a period has a high average low to close % of the day's range, it's indicative of a bull market.

More to come soon.

This is conceptually similar to the DV2, which uses a range calculation (over 2 days).

Thanks Scrilla

Indeed I doubt it is wholly original calculation. Something so simple is bound to have been applied in a similar way before. I suppose in this case I’m looking to isolate the low to close element of the range.

Originally I was more interested using the distribution of days having such strong intraday action as an average, which I’ll look at in a follow up post. I actually found it to be a good filter for the DV2, especially for the shorts which is what I was looking for.

Dave

You say that “Arguably 50% should be the cutoff”

Do you mean to avoid curve-fitting?

There’s a “mound of toast” around 50% +/- 5% would you say?

Thanks, and keep blogging!

Yes indeed to avoid curve fitting. There’s not too much difference in the results though.

Thanks for your comment

Dave