We started providing high probability trading setups in premium articles for educational purposes only in the report for week of April 25, 2016. In this article we provide details of the performance of these discretionary trading setups and compare it to that of a quantitative mean-reversion algo in the same period.
Among other many known and unknown factors, trading performance depends on risk and money management. Any claims of performance that do not reveal specific risk and money management details should be carefully scrutinized.
There are many stock signals providers in the blogosphere that calculate performance in ways that reveal that these individuals have little or even nothing to do with the business of trading.
For example, some of these individuals will provide a large number of setups during a certain period of time hoping that one of them will hit the jackpot. Then, they irresponsibly compound gains to generate unrealistic return figures. But in reality, traders have limited capital and they need to have an allocation scheme with specific risk levels. Under the conditions of realistic trading, there is a limit on open positions in order to maximize returns and capital may not be available to take that theoretically good trade that positively skews the results. We have discussed these facts in another article and have provided some examples of performance.
In real trading one has a C amount of capital (before or after leverage) and a universe of N securities. This is a difficult problem to solve because both position size of a single trade and also the number of trades must be maximized. There is no optimal solution because performance depends on future path that is unknown. Most experienced traders set the parameter in advance hoping that the solution is at least sub optimal.
We choose N = 3 and maximum risk per position as a percentage of initial capital C of 2%/N. Our universe is comprised of all Dow-30 stocks and 30 popular ETFs. We hold up to three open positions at any given time and we size then so that risk does not exceed 2%/3 or about 0.67%. This presupposes the following:
(1) There is an exact trade entry level and exact stop-loss level
(2) The levels in (1) will not differ substantially from actual entry and stop-loss
Condition (1) is needed to calculate the number of shares. Furthermore, if the number of shares results in large allocation than C/3, then it is adjusted accordingly.
Condition (2) is needed to minimize slippage. In the case of liquid securities slippage can be kept low, usually less than 50 – 100 basis points total per year. But that can vary depending on market conditions.
The above are required for practical short-term trading, yet they are the least discussed issues in financial blogosphere, maybe due to the fact that very few are actually involved with discretionary trading and most are oblivious of these issues.
After the long and boring introduction for those who are actively trading and have skin-in-the-game, we can now present the performance of the discretionary high probability setups below:
The number of triggered trades is 28 in the period shown. Win rate is 60.7%, profit factor is 1.43 and net return is 2.61%. Annualized net return is about 4.83% but this number rarely makes sense.
It may be seen that performance peaked after 17 trades at about 5% (mid July) and then there was a significant drawdown. It has been exceptionally hard to identify high reward:risk setups since July of this year. This may be temporary or even a permanent situation. Machine learning and algo trading are transforming the markets with traditional methods showing high failure rates. Many traders have found shelter in online platforms that offer data and tools for developing algos and using machine learning but there are obvious risks.
Other, more sophisticated traders have managed to break the barrier and move forward. We are proud that we have customers who have achieved that with a little help from our software tools but primarily because they are very intelligent and have strong quantitative background. For example, we have a customer in Europe who uses a previous version of our DLPAL software to calculate features for machine learning. He does that for about 50 stocks and trains Support Vector Machines. Then he scores new data and selects top stocks to go long and bottom stocks to go short.
Next, we compare the performance of the discretionary high probability setup to that of the PSI5 proprietary mean reversion algo in the same period. This algo trades long-only SPY on EOD data.
Net return is about 8% with same commission level, win rate is 90% and maximum drawdown is less than 3%. Profit factor is about 3.55.
Is there any reason to even consider discretionary high probability setups in a world dominated by algos? One can see in Twitter the last discretionary traders struggling and some rushing to delete tweets after the setup they have announced fail.
Adaptation is important because it is part of selection process in life and business. Those who fail to adapt will face the consequences. This is the philosophical part with real implications. As far as premium articles, we have decided not to accept any new subscribers after December 3, next month, and then transition the content slowly to quantitative trading setups. But that cannot be on a weekly basis so we must think of a solution to that. An interim solution was provided in terms of our mean-reversion signal service, which includes signals from the PSI5 algo mentioned above.
If you have any questions or comments, happy to connect on Twitter: @mikeharrisNY
Charting and backtesting program: Amibroker
Technical and quantitative analysis of Dow-30 stocks and 30 popular ETFs is included in our Weekly Premium Report. Market signals for longer-term traders are offered by our premium Market Signals service. Mean-reversion signals for short-term SPY traders are provided in our Mean Reversion report.