Once a hypothesis has successfully transitioned through the development and evaluation process to emerge as a tradable strategy, it is now time to deploy it: that is, to trade it “live”. Most traders do this in two phases – first by forward walking the strategy for a period of time, and only later by trading it with real money.
A walk-forward involves paper-trading a strategy to ensure that it behaves as planned and that all of the complexities associated with trading it are fully understood before committing real capital. Some traders use long walk-forward periods (6 months or more), others opt for shorter periods.
If the strategy is destined to be traded automatically, this is the time when the code and parameters associated with auto-trading are finalized and tested. Full-auto trading is extremely complex to implement and a number of unexpected issues often arise. It is therefore essential to test and retest the code, particularly if several potentially conflicting strategies are to be run concurrently. Note that many automatic systems are best described as “semi-automatic” since they actually involve some degree of user input.
Determining the amount of trading capital a strategy will use is typically done during the risk-management phase of the development process. The single most important objective of any trading or investment strategy is the preservation of capital, so determining the correct risk-adjusted position size for each strategy and/or each financial instrument is essential. A strategy prone to sizeable drawdowns, for example, will likely warrant the use of a relatively small position size. The same is the case if the strategy is to be applied to a highly volatile instrument.
Most traders usually start small during the early stages of the deployment process and slowly ramp up position size as they gain confidence in the strategy and the procedures associated with running it.
Position size is a particularly tricky issue for those trading several strategies concurrently. Each strategy is competing for money from the same limited pool of trading capital, making the task of allocating capital rather complex. Moreover, several similar strategies may trigger at the same time, with a similar long/short bias, thereby adding to the cumulative directional risk of the overall system. Ideally, capital should be allocated equally across several uncorrelated markets and strategies, but this is not always possible to do in the real world.
After paper-trading the strategy and ensuring that the full-auto or semi-auto trading system behaves as planned, the final step if of course to apply the strategy to a live trading account. Again, some may wish to start small, others – possibly those more confident in the integrity of their development and testing process – may opt to immediately use full position size.
It is very tempting for a trader to second guess a strategy, particularly one he hasn’t developed himself. Changing entry points, exit points, stops, position size, etc – all these actions, driven by greed and fear, are symptoms of either a lack of experience or of a lack of confidence in the strategy itself.
Confidence in a strategy can be achieved by adhering to a strict strategy development process. If the strategy was not developed in-house, then a comprehensive evaluation process will allow the trader to gain sufficient confidence to trade it consistently and dispassionately.
The key is of course to “plan the trade, trade the plan“. There are two reasons for this. First, one assumes that the strategy was diligently developed and optimized to generate the highest risk-adjusted profit expectancy. So deviating from the strategy will very likely “un-optimize” it and lead to diminished returns. Secondly, the only way to assess a strategy’s performance is to implement it exactly as it was designed. Changing parameters will make it impossible to objectively compare the strategy’s real-life performance against those of its backtests and forward-tests.
Execution & System Errors
These are inevitable and are almost never to the trader’s advantage. They can range from “fat-finger” trades (wrong size, wrong symbol, wrong entry/exit price) to system issues (computer problems, network outage, etc). Again, a systematic approach to trading is essential to keep these to a minimum.
The performance of a trading strategy should be reviewed on a regular basis, typically once a quarter. The objectives of these reviews are to:
a) Compare the live trading performance to the performance of the backtests/forward-test
This helps gauge the continued robustness of the strategy. It indicates whether the trading edge uncovered during the development phase is still strong in the current market environment. A fading edge could possibly lead to the strategy being suspended or ultimately retired.
b) Compare live trading performance to hypothetical trading performance
This serves to gauge how well the strategy is actually being implemented. Actual and hypothetical live trading results should in theory be identical. Differing results may indicate that certain strategy rules were not strictly adhered to (entries, exits, etc) or that some of the assumptions made when developing the strategy were erroneous (commissions, slippage, etc).
The results of the review process will lead a strategy to be maintained, suspended or retired from the production environment. Moreover, its good/poor performance may result in a change in its ranking within the trader’s portfolio of strategies, thereby impacting the capital that will be allocated to that specific strategy going forward.
Statistical edges come and go. Some get weaker over time, others get stronger. Some persist for several decades, others last just a few years and then flatten out or even switch bias.
The two charts above show examples of this behavior. On the left we have a seasonal system involving holding a long position in the SPX for 24 hours on the last trading day of the year (1961-2005). This simple strategy would have proved profitable 77% of the time until 1991. Then, without warning, the strategy turned unprofitable for 10 out of the following 14 years.
The equity curve on the right shows the result of buying a ten day low close in the S&P500 and holding the position for exactly for 2 days (1970 to 2015). As we can see, in the 1970s and much of the 1980s, prices had a strong tendency to “follow-through” on 10 day lows – i.e. to continue falling. From 1987 to 2015 the exact opposite appears to happen: a 10 day low is generally followed by a quick bounce, or reversion. And the inflection point of the equity curve occurred on a specific date, October 19 1987, also known as Black Monday.
There are many interesting reasons for this dramatic change in market behavior – from continuation to reversion – but the important point is that market biases and their associated trading edges can and do change over time, so tracking the robustness of each and every strategy in a trader’s portfolio is essential for the continued success of an overall trading system.