The platform guide users through the process, which includes:
1.Strategy Name and Description - the first step is to choose a name and provide a description of the strategy. Given that building the strategy is likely to be a test-and-fail iterative process with a few attempts before finding the optimal solution, it's important that the name and description make quick reference to the test’s version and a short description of the variables used;
2.Target Universe - investors can choose to focus on names included in (a) the S&P500 index, which is a list of the 500 largest, well-known names listed in the US that operate in a wide spectrum of sectors and indices; (b) the Nasdaq100, which are the 100 larger names listed in teh Nasdaq exchange that tilts heavily on tech and media sector; and, (c) the Limex Expanded Universe, which is a list of approximately 2,700 names listed in the US spanning a wider range of market caps and sectors. The choice of universe will affect the learning time required by the machine, as well as the weekly ranking and portfolio updates;
3.Portfolio Preferences - this step is where investors feed their portfolio criteria to the machine.
The first two steps (Long Positions and Short Positions) refer to their long and short exposures:
- Percentage to Long/Percentage to Short - investors tell the machine which portion of the ranking to use when selecting positions. For example, if an investor indicates “use 30% for the long and 20% for the short” the machine will look at the top-30% of the ranking to select stocks to buy long, and the bottom-20% to select stocks to sell short. This step is more relevant when using the S&P500 and Limex Expanded Universe targets as the analysis can be highly time-consuming if the machine needs to characterize each one of the stock’s risk-adjusted reward operating in tandem with the other members in the portfolio;
- Min/Max number of Longs/Shorts - what is the maximum and minimum number of names to be included on either side of the book (long side or short side).
- Min/Max Position per Stock - a complement of the prior step, what is the minimum and maximum size of each position on either side of the book.
- Portfolio Long/Short Percentage - what is the size of each side of the book compared to the value invested. This number can be between 0% and 300% to account for the use of leverage (up to a 2x debt-to-equity ratio). Investors that want to look at a pure long-only strategy will include 0% on the short side, and for using no leverage the size of the long book would be 100%.
Following the general Long/Short requirements, the user will identify macro variables for the portfolio including:
- Investment Horizon - the machine will follow different signals for investors with different expected holding periods. The default is 21 days, which is equivalent to a month in trading days.The machine will re-evaluate the portfolio on a weekly basis which means that a particular stock could have ranked highly on day 1 and poorly on day 8 (for example, if it rallied in the first week) so the machine will suggest to sell the position on the first refreshment.Importantly, signals that are created for 21-days expected holding periods tend to decay very slowly for longer horizons thus a reasonable rule of thumb is to keep the 21-day default if the user is a long-term holder. For momentum traders, shorter periods can have material differences.
- Portfolio Starting Value - this is a yardstick for the machine to put a monetary value on the performance calculations. The default is $100m, which tells the machine to start the back-testing simulation as if it had invested $100m on the first day and never redeemed capital.
- Trading Costs, Smooth Periods, Execution Delays - general execution conditions. The trading costs (default is 0.01%) should reflect all monetary fees (these days mostly at zero) and bid/ask spread inefficiencies. Smooth Periods (default is 1 day) is a dampener for signals that have higher daily volatility; it mostly has “bite” for stocks that are very illiquid and inefficiently traded, which could result in an unnecessarily high trading frequency and jumps in rankings. Execution Delay (default is 1 day) affects mostly the simulation as investors will likely trade on their own schedule - if investors expect to delay the execution of the trades by a day or two since receiving the portfolio and ranking suggestions from the machine, they can ask the platform to use the similar delay in their back-testing.
The next step is for the platform to ask whether to use Stop Loss/Take Profit rules. These are pretty standard:
- activation toggle switch - the first step is to activate (or not) each rule with the toggle switch at the bottom.
- Threshold and Percentage - the Threshold is the level of loss or profit to accept prior to triggering the stop loss/take profit action. The Percentage is the portion of the position to close when that action is taken.
- For example, a stop loss with a 10% Threshold and 50% Percentage to Sell implies that the machine will sell 50% of the position when the loss hits 10%.
Finally, the machine shows a number of toggle switches to activate or de-activate a few more variables ad-hoc to each portfolio description;
4.Optimization Objective - this is where users feed the machine what to prioritize, including:
- Optimization Goal - the user has to pick between maximizing the expected outperformance versus the wider market, minimizing the maximum drawdown (represented by the Value at Risk, VaR), or look for a combined risk/reward optimization maximizing the Sharpe ratio (the Sharpe ratio compares a portfolio’s “units” of additional return over an index - in this case the S&P500 - against the “units” of additional volatility; e.g., a Sharpe above 1 means the the incremental return is relatively higher than the incremental volatility).
- Lower Turnover, Optimize Turnover, Tracking Error Optimizer - these are pure portfolio management and trading rules to avoid excessive churn and aggressive deviations from the benchmark.