2 years of trackrecord for our darwins
ATR & PQH.
In spite of the results could seem good, we have crossed difficult problems with these darwins, related overall with a not good Risk Stability Rs, that implies a not good replication of the Risk Manager.
To understand what is the problem, why it is created, and how I have tried to solve it, I am posting again the explanation of the system:
The PQHANTITRENDSWING & ANTITREDSWING strategies underlying Darwin PQH and ATR are antitrend swing trading systems traded in 2 different moments of the trend, one in the early symptoms of depletion of the trend (LOW system), and a second later when we already see the trend really exhausted (HIGH system), and the success rate is higher, so we increase the leverage of the trades and the ratio RR/trade in this type of setups.
As we can see, we have a multiplicator factor between the systems LOW & HIGH. Here is the key of the problem, because we can conclude that the factor chosen (aproximately 10) for the Darwin has been too many big to get a good Risk Stability and to get a good replication of the Risk Manager.
To understand better the concept, attached a chart with the daily lots of the backtest of the underlying strategy of ATR. We can also see drawn the moving average of the daily lots calculated in 45 days (as VAR), 9 months (3 DPeriod), 1,5 year (6 DPeriod) and 3 years (12 DPeriod).
As the var of the darwin is calculated with the last 45 days, we can asimilate that the orange curve would be similar to the VAR of the Darwin. We can see why the Risk Stability is not good.
When the period of calculation is higher, the Stability improves, then to get a better replication the period of calculation of the var would have to be higher.
All of this, implies that the Darwin is increasing the risk working with more leverage that it should be in the periods of low volatility (system LOW), and limiting the leverage in the periods of the system HIGH.
Then the performance of the Darwin was being lead for the performance of the LOW SYSTEM, working very good when LOW SYSTEM Works well as happened in 2016, but losing the compensation of the HIGH SYSTEM in the periods as 2017 when the LOW SYSTEM doesn't work so well.
They are two independent systems, but we have joined in the darwins for two reasons:
If we had only created the LOW system, there could be long periods of drawdown or stagnation, in moments of trend market volatility. This volatility generates more losers of the LOW system trades, but they give the choice to trade HIGH setups, and to improve the return of these periods. The fruit of this, it is the annual stability of the Profit Factor obtained in the backtest, avoiding the years where the LOW system behaves worse.
As the Darwin has worked as the LOW SYSTEM, 2017 suffered one of this periods of long stagnation without the effect of the HIGH SYSTEM.
Last summer I realised of this problem, when get to solve this problem was my first priority in the development of our darwins.
It is necessary to improve the Risk Stability Rs, and to get it it is necessary to reduce the amplitude of the chart of the lots, and get a more stable curve of the lots moving average.
In a first phase of improvement of the Darwins implemented last february was to reduce the multiplicator factor of leverage between LOW&HIGH SYSTEM. Now both darwins are working with 100% algorithmic trading, and the factor has been reduced for ATR to aprox. 6, and 3 for PQH.
Obviously when it drops the factor the amplitude of the moving averages is reduced, and the stability is better, as we can see in the charts attached of ATR & PQH.
In the new configuration of ATR the maximun daily lots have dropped from 1,68 lots to 1,0 lot; while the amplitude of the moving average of 45 days has been also reduced from an average of 0,16 to a maximun of 1,0 lot, to an average of 0,15 to a maximun of 0,55 (more than 50% of amplitude reduction)
In the new configuration of PQH the maximun daily lots have also dropped to 1,0 lot; while the amplitude of the moving average of 45 days has been also reduced from an average of 0,20 to a maximun of 0,5 lot (almost 65% of reduction from old configuration, and 25% more from ATR configuration).
These new configurations Will improve the replication, and consequently the score of the IA Rs.
Not only, we have changed the multiplicator factor, we have discarded some sets in some pairs that they were not working as we hoped, and we also have improved the Market correlation of the Darwin (score that we Will see improve smoothly), because the sets of SELL and BUY were not well balanced. Now, we have balanced them in number of trades, and exposition.