Machine Learning References and Backtesting Checks for Trading Strategies
Summary
The document collects suggestions for learning about machine learning and data mining in quantitative investing, including a standard statistical learning textbook. One response treats data mining and machine learning as closely related terms. Another recommends investigating seasonal and calendar patterns, while emphasizing that any proposed strategy needs independent backtesting because market behavior changes over time.
That response lists practical diagnostics for strategy evaluation, including trade counts, frequency, long and short balance, win rate, reward relative to risk, expectancy, time in trades, drawdown, and risk per trade. It also recommends testing across in-sample and out-of-sample periods to reduce bias. These are screening and validation ideas rather than a complete research protocol: the document gives no test results, dataset, or detailed procedures for preventing leakage, multiple-testing bias, or overfitting. Its claim that traders rarely use machine learning is anecdotal, so it should not be read as evidence about industry practice.
Key ideas
- A statistical learning textbook is suggested as a starting point for machine learning study.
- The responses describe data mining and machine learning as closely related fields.
- Seasonal and calendar patterns are offered as possible subjects for trading research.
- Backtests should examine trade counts, performance, exposure, and drawdowns.
- Testing across separate in-sample and out-of-sample periods can help assess whether a pattern generalizes.
Tags
Full text
# Buy side techniques # Buy side techniques I was speaking with a friend of mine about what techniques are used for quantitative investment management, and he told me that, when assuming active positions on the market, even in high-frequency trading, the most valuable knowledge is about time series analysis, data mining and machine learning. Now, apart from time series analysis, I don't know well the other two aforementioned fields. Can anybody suggest references for getting better in those areas? ## Answer by madilyn (score 2) https://quant.stackexchange.com/a/21044 The classic text for machine learning is 'The Elements of Statistical Learning' by Tibshirani et al. I believe the term "data mining" is often used synonymously with "machine learning". ## Answer by paglos (score 1) https://quant.stackexchange.com/a/21053 I second Tibshirani's book. There is an another edition you can download free on internet : http://www-bcf.usc.edu/~gareth/ISL/ ## Answer by Koray Güclü (score 0) https://quant.stackexchange.com/a/21066 Buy side techniques such as machine learning might be useful but i have not seen any trader applying this method. (see market wizards book series) Some tried but stopped using. Larry Williams, and seasonal, cyclical traders are using patterns based on month of day, days of week etc... you can analyze them as well. There is also best know 6 months of year pattern (see Traders Almanac) You must know how to backtest any trading/investing strategy before applying them even you read them in a book from a master trader. You have to analyse the strategy by yourself. Do not forget that the markets do not behave the same as they where before. Some of the key metrics to analyse any daily pattern or trading/investing strategy are: - No Trades (must be greater than 30 – pref. much more) - No. Trades/ Year/ Month (whatever time period is relevant) - % Long Trades/ % Short Trades (is there a bias?) - Win%/ Success Rate - Average R:R - Expectancy (Average profit per dollar risked) – Must be a positive number! - Avg. no days in trade (Carry costs/ duration exposed to risk) - Maximum R multiple drawdown measured - Min/ Max/ Avg. Pips Risked (see http://blog.cmcmarkets.com.au/analysts/ric-spooner/backtesting-trading-strategies/) To avoid any biases. You have to test your method by using different time periods such as insample outof sample etc... to validate its validity. blindly apply datamining, or machine learning will not guide you anywhere.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.