Portfolio Diversification and Risk Management
The benefit of using software to develop or generate strategies is that we ended up with tons of profitable strategies and we are left with the decision of how to manage these hundreds of strategies, which to choose, how to choose them.
A) Risk Management - the use of MAR and Sortino Ratio
In term of risk management, the MAR and Sortino Ratio are recommended because their take into account of the risk exposure of the strategy. Though the Sharpe ratio has it's benefit, the MAR and Sortino Ratio put more emphasis on the downside standard deviation, ie. max DD. The upside standard deviation doesn't pose any threat to the portfolio risk exposure and hence, the MAR and Sortino Ratio are more favored as the metric to use.
In order to diversify our portfolio, there at least 2 types of methods.
1. Diversification via different Markets/Symbols, and
2. Diversification via different strategies (namely Divergent Strategies and Convergent Strategies)
B) Diversification via Different Markets
Don't put all your eggs in one basket. Hence, when choosing different markets to trade, pay attention to the correlation metrics of the currency pairs.
Basically, the currencies are grouped into 2 different types
a) Consumer Driven - such as currencies traded in USD, JPY, CHF
b) Commodity Driven - AUD, NZD, CAD, GBP
Thus, if you are going to diversify, choose pairs that are in these 2 different baskets, otherwise, they are essentially the same (albeit, varied volatility)
C) Diversification via Different Strategies
Basically, most of us are either Breakout Traders (Divergent Strategies) or Range/Mean Reversion Traders (Convergent Strategies). These two approaches are mirror opposite in their outcomes.
A convergent strategist believes that fundamental value approaches can be used to get an idea of the intrinsic value of an asset. The price will eventually converge to this intrinsic value, i.e. return to the mean. The convergent strategy has many small gains and catastrophic losses.
A divergent strategist believes that past patterns can predict future patterns. He believes that past patterns are due to changing and adaptive actions by investors. These reactions are driven by investors’ emotions and irrational factor.
In addition, a lot of factors non-quantitative effects due to government intervention, war or politics, are not quantifiable in fundamental analysis methods. Furthermore factors of adaptation, as well as behavioural biases, may influence prices which differ substantially from their intrinsic value and hence cause breakout. The divergent strategy has many small losses and euphoric wins. And it's weakness is, if there is a prolong period of ranging, the portfolio would be at a risk.
Though I tend to favor Divergent Strategies (cos I'm a breakout trader), most strategies—and investors’ portfolios, for
that matter—are convergent-focused. Reason being, the market is ranging 75% to 80% of the time and thus the saying goes, make hay while the sun shines.
However, the market is irrespective of what you like or favor (whether you are a divergent or convergent trader). The winner is the one who is most adaptive to what the market dictates. Hence, balancing our portfolio with both Divergent and Convergent Strategies, is another form of risk management one need to consider during strategies selection for a portfolio.
Generator Bias?
a) Win/Loss Ratio Bias
If we were to understand the characteristics of Divergent and Convergent Strategies, we also realised that Convergent Strategies tend to be the easiest one to be identified/searched/generated.
A Convergent Strategies have many wins and a few catastrophic losses (due to breakouts period) and hence, if one were to use the Win/Loss Ratio, one would most likely ended up with Convergent Strategies.
I propose that we have another type of Acceptance Criteria. That is Maximum Win/Loss Ratio. Whereby we want to search for strategies with Win/Loss ratio not more than 50% or 35%. Divergent Strategies have lower Win/Loss Ratio and they tend to make many small losses in between the Breakouts, hence limiting the Maximum Win/loss ratio, we prevented the Generator to search for Convergent Strategies that tend to have higher Win/loss ratio.
b) Time Period/Data Horizon Bias
Since the market is ranging 75% to 80% of the time, at any given starting point, Convergent Strategies would be the one being identified as winning strategies (till a breakout come along) and hence, I'm curious whether the Generator somehow would be bias towards generating Convergent Strategies because once you click the Start button and the generator run through the data horizon (time line), which types of strategies would start making money first?
Therefore, it's no wonder whenever I run the Generator in both FSB Pro and EAS, I tend to have Convergent Strategies in my collection. Often when I examine the trading rules of these Convergent Strategies, they contain (anti trending rules, ie. mean reversion rules). Part of the reason why I don't really use the Generator (in the past) is that 1) I'm a breakout trader and 2) I don't believe that convergent strategies are good for the long run...you need only a black swan to one day blow up the account...which we seen many portfolio blew up during such events.
EA Studio has a max of 200k bar and if we were to use any time chart below H1, we would have very short period of data horizon and thus I believe with such short data horizon, the Generator may be exposed to Convergent Bias - has the tendency to generate Convergent Strategies (because how many black swan events occur within 3-4 yrs?). If we want a more robust strategy/portfolio, we don't just hope such black swan event don't occur, we are always prepared for the worse. Hence, convergent strategies do look better than Divergent strategies in the short run but if we were to examine them in a longer scope of time, the Divergent strategies are the one, most likely stand the test of time.
I once suggested to remove anti trending rules in the Generator, i.e do not use rules such as Falling, below than, change direction down, cross downward etc in Long/Buy position and vice versa, do not use rules such as Rising, higher than, cross over and change direction up for Short/Sell positions.
However, on hindsight, what I suggested would then make the Generator to be Divergent Bias, i.e. search for Divergent/Breakout/Trending Strategies.
Since we need to ensure our portfolio contain both Divergent and Convergent Strategies. It would be good, we have greater control of the types of strategies we generate.
Hence I suggest that we have the option to check on a box, "Trending Strategies Only" or "Divergent Strategies Only" and "Convergent Strategies Only" whereby trending rules are eliminated in the search.
In another words, I hope Popov could add in the generator algorithm to search only trending rules if we tick the box "Divergent Strategies Only". And the algorithm for the Generator would use only trending rules. That would cut down the possible search number by half and speed up our process of finding Divergent/Trending/Breakout Strategies (cos the other half of the options such as Falling, cross downward, lower than etc would not be use at all in the search of Long/Buy rules).
Convergent and Divergent Strategies are exact mirror of each other in terms of
a) Trading rules, one uses mean reversion rules and the other uses trending rules,
b) in terms of trading period, one trade during ranging and calm market and another trade during breakout and volatile market.
The whole idea of using Correlation Metric is to prevent our generated strategies in the Collection ending up with similar strategies.
With the option of switching in between Convergent and Divergent Search/Generation (by removing either convergent rules or divergent rules from being generated), we would have a broader spectrum of different types of strategies that are non correlated. I believe this method of sieving out strategies by trading rules would be a more efficient method than simply using correlation metric because we go straight to the source/cause of why strategies are not correlated, due to rules and not simply some statistics differences. It's the strategies rules that contributed to the statistics but statistics don't make the rules, it's the rule that make the statistics cos we are measuring the strategy performance outcome based on their trading rules.
In conclusion, I propose that Generator (both FSB Pro and EAS) has the added function to opt out certain types of trading rules in our search and the option to choose when to turn on/off this function so that we are in greater control of the Generated strategies outcome/quality/types. By clicking on the option to choose between generating Trending rules or not, I can therefore deliberately create both Convergent and Divergent strategies in my portfolio. Currently, most of the time, I ended up mostly with Convergent strategies due to the possible Generator Bias as I mentioned above.
Note: To read more about this option to turn on/off, read this post I put up "Increase Generator Success Rate"
https://forexsb.com/forum/topic/7293/in … cess-rate/