Exploiting inefficiencies after large downswings

[nick normandin]

Hello,

I'm looking for any relevant literature on strategies that involve selecting assets that have recently experienced big downswings. I remember reading something in Fortune's Formula about an investor (I think it was Claude Shannon??) making money from a strategy like this, but it was short-lived and not scaleable to be used by a large fund. The underlying theory was that panic sets in when a stock starts plummeting, which results in an inefficient price that then might experience some form of correction.

Before any EMH enthusiasts jump down my throat, this is for purely academic research purposes. I would appreciate any papers or leads. Thanks!

Nick

 

[Harkkam]

Well how would you first classify when an asset has become miss priced? And two what time frame of mis pricing are you looking to quantify.

Just because there is panic in a stock does not mean it is mis priced, for example breaking news about a lawsuit or a failed drug can significantly change the underlying fundamentals, so even though you think you are buying a "Dip" is really a fundamental shift in value. So not all sell offs are just panic.

So you would need an algo that will look at the current price of a stock and then screen out for fundamental news and you would need to design an algo that can read that news and automatically judge the seriousness of that news. Say your algo decides that breaking news on this stock is a level 4 event which you have defined yourself and statistically all of your defined level 4 events have resulted in a intraday 5 minute time frame movement of about .5-1%.

Your algo will look for deviations from this then, say stock xyz reports a level 4 news event except it has fallen say for example 3% which is almost 2 STD's from your defined .5-1% change in 5 minutes.

That stock XYZ's move on level 4 news now may be an anomaly, 3-5% vs .5-1%.

This stock XYZ is now a likely candidate.

This is very rudimentary, but like that there are hundreds of pieces of information, but the general idea is to have a base line and look for deviations from that base line, using statistics to back up your strategies.

 

[nick normandin]

First, thanks for your response.

I believe that what you're getting at is incorporating text analysis into the algorithm? That would indeed by ideal, but due to time constraints my idea was actually a bit simpler. I'm planning on feeding in data from stocks that have taken big hits, calculate a bunch of technical analysis/volatility/other indicators that I made up, pass it through a correlation based feature selection filter, and then see if some machine learning algorithms can recognize any patterns across various time horizons. I'm using daily returns for one major index to train the algorithm and (if I have time) testing the trained classifier on daily returns form another index to analyze the robustness of my findings.

My research isn't trying to show that every security that suffers a large downswing (more than 10%, let's say) is mispriced.. just that sometimes there is an overreaction (which is then corrected for), and maybe there are common traits among those events. I'm really just looking for any information about times that this strategy has been implemented in the past.

 

[Lyosha]

Enron = best mean reversion candidate ever. And lets not forget the Brent - WTI spread trade! ... and all those nice Tech companies during the Clinton years...

All times when your strategy has been implemented in the past (and caused some hair loss....)