http://www.financialmarketresearch.net/

This problem really has two aspects; determining the market environment (bull or bear), and then deriving factors to gain the maximum RAR from that environment.

IMO the market environment question really comes down to market timing (for example, it would be useless to measure price v. the 200 day ma unless this exercise had some predictive value).

The question is then what is the most effective market timing model. Academic research has determined pretty conclusively that stocks are positively autocorrelated in the 3-12 month horizon, and negatively autocorrelated in the 1 month and under horizon. For a basic market timing model, it is then a matter of selecting and testing a two factor model which selects one longer term and one shorter term factor (moving average, ROC, advance/decline measures, etc). The more time one applies to researching and backtesting the factors, the better the model.

The really interesting aspect of market timing, and the segway into the response to your question, is when one combines sentiment data into the market timing and factor selection process.

Denys Glushkov from Barclays demonstrates pretty well in the paper below that sentiment can not only help with the market timing problem, but in the question of factor selection as well.

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=862444

I think if one expirements with the momentum/mean-reversion technique I discussed above, and combines that with sentiment data, a robust market timing model will be the result (for an example, the blog regimenia demonstrates an excellent fusion of momentum and sentiment models into an asset allocation model).

Then, finding inspiration from Glushkov, one can construct a high-beta portfolio for the times when the model is positive, and a low-beta portfolio when the model is negative (of course the factor doesn’t have to be Beta, but it is something that performs well in my testing).

Scoring algorithm is here:
http://www.billakanodoodahs.com/2008/01/trpits-part-iii/

To some extent, taking what works long (espec. in bull markets) works really well when you flip it in a bear market.

If you have monthly output for test long and short data, you can always segregate it by whether the 100 ema was above the 180 ema for the S&P 500 (or in your case, the S&P 500 was above the 12-month sma).

Or, you could simply take a long-only set of factors that develop a portfolio with high beta (I mean REAL beta, as in slope of regression line beta), and go short those stocks in bear markets (assuming you have a good definition of those) and go long them in bull markets.

I’m not aware of any academic work on the subject.

Here’s another question, or really, a statement. Looking for factors that work in down markets better than in up markets presupposes that the modeler’s universe is domestic stocks and the benchmark is a relative one …

bill – still doesn’t answer what i am looking for, but thanks for the note.

-Gary

Bill – Yeah I’ve read O’Neils books, but that doesn’t help when market is going down. . .looking for predictive factors when that happens as well.

They called it “CAN-something” or other …

Verily, I say to thee, there is nothing new under the sun.