Episode 52: How to develop quantitative trading strategies - Ernest P. Chan Link to heading

• Data in finance is sparse, as compared to other domains, e.g trading data of Daily frequency is limited in backtesting, hence you need market intuition.
• Intra-day trading, is useful since it presents many trading opportunity i.e. Trades are held for hours.
• Seperate asset classes, and even securities could demand different models.
• Run stocks in market neutral strategies.
• At times, in strategies which have not taken into account all factors, it is important to override them, e.g. during natural disasters, major economic events, etc.

Diversification Link to heading

Portfolio face different risk factors, and a way to deal with that is to diversify. Strategies face risks such as:

• Long only: Market neutral.
• Market-Neutral: Sectoral risk, risk of exposure to certain sectors.

Every strategy has sensitivity to different risk. The goal of diversification is to neutralize risk in our portfolio e.g. market-neutral portfolios generate returns uncorrelated to market.

If you have one strategy only, you can only trade it, but check the leverage.

Regime-shifting models - Models which are able to capture momentum, and mean-reversion automatically i.e. Model predicts whether next month is a momentum, or mean reversion strategy.

Capital allocation Link to heading

• Capital allocation can be done after some time-in-market. With new trades, you cannot really know how much to allocate.
• You can’t always know market impact unless you use substantial amount.
• If you have a new strategy, it is advisable to start with little capital.

Characteristics of mean-reversion trades:

• Almost every strategy trades are similar in returns
• They rarely make huge losses.

Characteristics of mean-reversion trades

• A lot of trades might not be successful.
• Returns are pareto distributed: 80% of returns are due to 20% of trades.

Generating trading strategies Link to heading

• Books on trading.
• Articles, blogs on trading.
• Podcasts.
• From live trading, some insight comes to your brain, when real money is at stake, that you would never ben able to capture when its just a backtest. Some times, there are unusual patterns in PnL, which can be utilized in develooing new trading models.
• Academic research papers.

Whatever strategies you read should not be used as a recipe, but only as inspiration. Always make modifications to the strategy.

• When the strategy (which previously performed well), stops performing, you should diagnose it, to know when it stops making money, and make modifications to the strategy.
• Modifications might necessarily improve return, but it might lower the risk

Backtesting Link to heading

Pitfalls in backtesting:

• Over fitting: Model cannot generalize well to future data. A cure to that is to ensure the model is parsimonious and as sensible as possible. The sharpe ratio is a good metric of statistical significance of a strategy. A high sharpe ratio shows a good performing strategy. A strategy with high sharpe ratio is expected to perform well on live market. One can easily tell if the backtest is valid or not, since if the strategy is not able to replicate the backtest, then the backtest was not realistic.

Misconceptions of quant trading Link to heading

To get started:

• Start implementing some simple strategies which you learn from magazines, blogs etc.
• Learn some programming language.(Python, R or Matlab)
• Continue implementing strategies.