Thought I'd share some things I'd developed over the years. I would love feedback or new ideas.

1. Signal strength is correlated with forward looking profitability. If you bucket your signal strength into 10 buckets, bucket 10 should have a higher sharpe ratio than bucket 1 out of sample.

2. Strategy PNL is sufficiently autocorrelated that introducing stop losses does not destroy the overall Sharpe ratio.

3. Choosing a rolling optimum window on a lagged basis does not damage strategy returns (or even improves them).

As a corollary, if there is alpha in the short term (10-21 day) rolling worst window it is a bad sign because it means that other people are implementing your strategy, stopping out and providing liquidity.

4. The data necessary to execute the strategy is hard to collect. Assuming Bridgewater hiring machine learning arbs all easily detectable statistical anomalies in medium term.

5. The data predicts other things that could signal equity improvements. Example: if you're predicting shipment volume using letter of credit issuance and that has a feedback effect into shipping stocks, you also want to see your LC data predict shipping volumes not just stock prices.

6. There can be a clear 'liquidity provider' identified. Whether retail, central banks, taxpayers, hedgers, distressed companies etc. Insofar as you're not accessing beta, probably 0 sum.

7. Long term signals are uncorrelated with 13-F holdings. If you're running a price/book model and realize that 8/10 hedge funds have holdings sized by price/book then you're in a crowded trade.

8. Volume is uncorrelated with signal entry and exit. Another way to detect crowded strategies.

9. Thesis scales to another asset market price action. i.e. you have something working that trades energy stocks, hopefully it works on oil as well. For equity fundamental factors, if there's no particular reason they shouldn't work in Japan, they should work in Japan.

10. Predicting a stationary timeseries is more valuable than predicting a trending one. Predicting the relationship between gold and gold miner stocks is more valuable than predicting the gold price. Predicting NOKSEK is more valuable than predicting EURUSD because it has shorter directional stretches.


WordPress database error: [Table './dailyspeculations_com_@002d_dailywordpress/wp_comments' is marked as crashed and last (automatic?) repair failed]
SELECT * FROM wp_comments WHERE comment_post_ID = '9245' AND comment_approved = '1' ORDER BY comment_date




Speak your mind


Resources & Links