## Friday, March 13, 2009

### Mean-reversion is getting stronger

As I mentioned in various previous blog posts, (e.g. see here), I believe mean-reversion strategies have been performing very well in the last year. Now here is an article (hat tip: Laurence) that provides concrete analysis to support this hypothesis. In fact, the author points out that most of the mean-reversion in recent years comes from the overnight close-to-open reversal.

Subscribe to:
Post Comments (Atom)

## 15 comments:

Dear Ernest,

I have your book and tried to login at epchan.com/book/sp500_20071121.xls but it failed. I used the correct password and login... Is it not available anymore?

Newton.

I posted a review of your book here http://blog.traderwerks.com/2009/03/01/quantitative-trading-el-book-review/

You may want to take a look.

Thanks, Trader, for the review!

Newton: the file is still available, but you need to select "Excel" as the application to open it when you click on the link.

Ernest,

Here is the message that appears:

Error 404 Page Not Found

Sorry - Error 404 | Page Not Found

Newton,

sp should be SP in the file name.

Ernie

Ok, Thanks!

It works now.

I trade mean reversion in the direction of the drift. This strategy has worked very well for me but this year has been an absolute blow out in terms of performance. We close our year next week and so far we are up 48%. It has been a long time since things have worked so well.

Proxy based spread strategies are also working phenomenally and are adding a lot of stability to the portfolio.

Anonymous,

Congrats to your great performance and thanks for sharing your results with us!

By "trading mean-reversion with a drift" do you mean you have an expected value of the drift? Do you compute that drift based on a technical or fundamental model?

Ernie

Ernie,

Can't really share our indicator or its construct but I can say that we trade mean reversion taking trend into consideration. The trend has obviously been down in the S&P500 over for the last few months and we have only been taking mean reverting trades to the downside (ie, selling rallies - although the AI and NN algorithms are more complex than that).

S -

I liked your book, but it is very annoying that the premium section of your website (http://epchan.com/book/) does not allow directory listings (it gives a 403 error). There is also no index file, despite what you say on http://www.epchan.com/subscription/spread.htm. So there is no way to know what files there are other than trying all the filenames mentioned in your book (some of which are case sensitive, others not). This also means that if you post a new file, we will never be able to find it.

HI Ernie,

Was looking for mean reversion trading over the longer term and came across your post.

Your post is relevant for reverting to the last trading days avg. price. But what about mean reversion over the longer time period ? For example reversion to the 200 DMA after being say +40% above the 200 DMA.

Need your views on this.

I have done a backtesting study on this on Indian NSE INDIA NIFTY index for the last 18 years. And based on that predicted a 15% fall in NIFTY by this month Nov`09 end. In fact I believe that most global markets have the above similar pattern. see details at my blog http://saanpaurseedi.blogspot.com

Cheers

~ The Fixer

Fixer,

Sure, there are mean-reversion at all different time frames. I focus on the short-term because short-term strategies have higher Sharpe ratio.

Ernie

Hey just did exercise 3.4 in book IGE

& got a sharpe ratio -.334220267

as opposed the proposed .789317538 in your book. What did I do wrong?

Hello Ernest!

I have a short question regarding your mean reversion calculations:

In Example 2.4 of Algorithmic Trading, you regress deltaY on ylag.

In Example 7.5 of Quantitative Trading, it is dz on prevz-mean(prevz).

Can you explain the difference why in one case the mean is used and in the other it is not?

Thanks!

Felix

Hi Felix,

The regression coefficient (slope) is actually unaffected whether you add the mean as an offset. In both cases, the offset is non-zero, and is itself a result of the regression fit. Adding the mean only affects the fitted offset, not the slope

Best,

Ernie

Post a Comment