In this post we explain the new formula of the D-Score, our flagship metric that certifies the quality of listed trading strategies and their associated DARWINs.
THE AIM OF THE D-SCORE
Some of the most frequently asked questions by traders, of us and themselves alike, are:
- How does Darwinex evaluate the quality of its own algorithms? and,
- What are the reasons that dictate the selection of one DARWIN over another.
To answer these effectively, it’s important to first recall the main aim of the D-Score.
The motivation behind the D-Score is to find the value that best predicts the chances of a DARWIN winning in the future.
Additionally, we also want this value to depend on some Investable Attributes, public and accessible for the trader.
This serves as a guide to orient traders on how to improve their score, which will result in better DARWINs over time, growth in investor capital and performance fees for traders, etc, creating a virtual circle endlessly feeding itself.
The following short tutorial presents the characteristics that any given trading parameter needs to exhibit, in order to be considered an Investable Attribute (this is a captions-only tutorial, please activate the subtitles in English).
WHY NOW? MORE QUALITY AND QUANTITY OF DATA
When it comes to affecting important changes in algorithms with the aim of improving their predictive power, it’s important to bear in mind one fundamental factor: the size of the dataset.
The less data used in a sample, the more random and unpredictable the final result will be, regardless of the DARWIN’s D-Score.
Our current sample of DARWINs is of much greater quality today, compared with those we had in the last iteration. At that time, approximately a year ago, we prioritized simplicity above all.
As a result, any solutions back then came from a sample that was not as significant as it is today. This concerns in particular, DARWINs with good scores, from whom definitive conclusions could not be drawn at the time.
It is important to remember that the old D-Score was calculated thanks to a combination of the 12 Investable Attributes, which always had the same weight in the final score, regardless of the inherent characteristics of the strategy behind the DARWIN.
In the following graph we show the backtest of portfolios with a VaR of 2.5% that filter daily in terms of a determined D-Score – old formula – (>10, >20, >30, >40, >50, >60, >65,>66, >67, >68, >69, >70, >71, etc).
The return that is shown in the vertical axis has to be multiplied by 100, so a 0.2 would be equal to 20% whereas a -0.8 is tantamount to -80%.
Up to a D-Score below 65 the results are as expected, because they produce returns which are gradually superior as the D-Score improves.
However, above 65 an erratic behaviour can be observed in the returns obtained, which is backed up by the current data and was what made us reflect on the suitability of the D-Score formula.
SEARCHING FOR THE PERFECT COMBINATION. A CHIMERA?
There are an infinite number of combinations in terms of the weighting of the 12 Investable Attributes which make up the D-Score.
The one which best predicts the chances of winning in the future would be the ideal combination, but does such a combination exist?
On one hand, if we create a portfolio of DARWINs which invests each day in terms of whether the DARWINs fulfill a determined minimum D-Score – what should be accomplished is that the higher the score, the better the future return of the portfolio, right?
On the other hand, it would be ideal if after a certain Score there would always be consistent positive returns, so that the guiding role of the D-Score makes sense for the traders.
As such, we have been working very hard to find the algorithmic combination that fulfills two conditions:
- Guides the traders, and
- Is predictive of future results.
NEW D-SCORE FORMULA
Findings from our research indicate that there is a pattern whereby there are minimum D-Scores which, occurring with certain Investable Attributes and combined with a good Performance (Pf) Score, offer better backtest results than other DARWINs which do not meet this minimum Score.
This means that a score of 8 in some Investable Attributes produces a much better result than if the score were 6.5 despite the fact that the difference of 1.5 points might seem insignificant.
On top of this, the new D-Score formula is going to take into account the time period of the strategy and will customize the weightings allocated to the Investable Attributes based on the strategy “timeframe”, since it makes sense that, for example, the Timing scores (Os/Cs) in short term strategies are more heavily weighted than in long term strategies where the La score ought to have more importance.
CONCLUSION: MORE QUALITY, MORE AUM QUANTITY
Following on from this, we can say that the new D-Score improves the average score of the DARWINs, as well as the performance of the backtests*.
And not only that, but with this improvement the Score increases, which leads us to believe that the new D-Score is a qualitative improvement on the old one.
Taking into consideration the remarkable improvement on how we rate trading strategies/DARWINs, DarwinexLabs is going to increase the amount allocated to the DARWINs in which we invest.
*Compared with the old calculations, the current return curves, and taking the same level of minimum D-Score, the performance lowers. This is due to the fact that they included more DARWINs in percentage terms. Nevertheless, if we carry out the comparison based on the same DARWINs, the new D-Score achieves better results.