The new Risk Manager 2.0 (Part 3)
This is the third post of our series of articles regarding the new Risk Manager. If you still have not read them, we recommend that you read the first post about the reduction of VaR from 20% to 10% and the second post in which we spoke about the changes in the calculation of the VaR.
The change that we are going to explain in this third article is the fruit of the feedback that we have received from a few traders in our community. During the last few years, we have received various reasonable complaints from traders that have criticised that a DARWIN has experienced losses when the underlying strategy, having a reasonably low and stable VaR , would generate positive returns.
In our continual drive to improve our system, we decided to analyze all of our strategies to see what the reasons were behind the phenomenon that our users had reported.
This analysis helped us to find an error in the approach at the time of defining the maximum permitted leverage for the underlying strategies, from which our risk manager closed the investor’s exposure in order to guarantee the 20% monthly VaR risk of the investors ( n.b: from the launch of Darwinex reloaded , the VaR objective for the investor will become 10%).
Just like we have explained in previous articles, the investor leverage (i.e of the DARWIN) is calculated from the following function:
Investor leverage = strategy leverage *VaR (10%) / VaR (strategy) * % risk adjusted for excessive leverage
The risk adjustment always reduces the percentage leverage in such a way that the DARWIN yields and those of the strategy are left to behave in a linear way.
Thus it is very important to be very rigorous in the calculation. Now we will proceed to the detailed explanation of the changes that we are going to introduce in the launch of the new risk manager.
- Why do we need the second component of the formula (the red component)
To explain the structure of the above formula, we are going to illustrate in an example. Let´s suppose that a strategy operates by always opening the same number of positions every month, 20 for example, with the same leverage, for example 10:1, and whose durations are always an hour. The VaR of this strategy we are going to assume is 8%.
What would occur if suddenly this strategy decided to leave it’s positions open for a longer time or even indefinitely. Obviously, if the investors maintained this leverage during the month, the investor’s risk would be higher than 8% monthly (with 300 pips of expected movement in a month investors could lose 30% of the capital, a lot higher than the 8% calculated). It is because of this, the time elapsed, that the risk manager would have to act to close a part of the investor’s exposure.
This example (and there are many more examples), serves to explain why the maximum defined leverage for a position must depend on the duration of the position, and therefore, the risk manager must be able to act along the entire life of the position, and NOT only at the opening. Every time that a position is opened, there are temporary windows of risk performance, and for each of them, there has to be a definition of the maximum tolerated leverage, which in addition, always must decrease as time passes.
- Position history
The maximum leverage of a position is going to depend on the type of underlying strategy, so that we must base ourselves on that history to fix it. Mathematically, the most sensible way is that the selected history to define the VaR of the DARWIN coincides with that selected to determine the VaR of the strategy, so that the adjustment must be connected with the VaR calculated before the strategy that you want to control is opened.
- Room for improvement in our closing algorithm
The fact that we actually were fulfilling the risk objective in the DARWINS, does not signify that we are optimally closing all of the positions in all of the DARWINS. This point was one that we didn’t notice when we initially designed our model.
Initially, we assumed that the fact that we were getting an objective VaR in the 20% DARWINS was because the adjustments that we were using at the end were correct. Time has shown us that, unfortunately, this assertion isn’t certain , and because of that, we have given it a “facelift” to optimize our criteria of closing positions.
If we deviate from our historical sample of “position duration” vs “leverage” of a strategy, and group them in unlimited ranges for those instances where their performance has been predetermined by the risk manager, we can obtain for each duration range the medium and dispersion of it’s leverage. We have attached a graphical explanation.
There is a maximum permitted leverage for the risk manager in terms of the “number of dispersal times” that has been taken into account in ALL the resultant DARWINS, the monthly profitability distribution works in that only 5% lost more than 20% (in the future we will work with a VaR of 10%).
In the previous example, in a timeframe of 4 to 8 hours, the maximum permitted value is 17.07 so that at 4 hours the risk manager would close whatever position whose leverage was higher than that at that time (red dots in the picture).
Essentially , thus we have fixed the closing criteria (as we have mentioned before, this solution is only one of thousands of possible options).
What can we improve?
Our initial solution is valid for our universe of aggregated DARWINS, but not for each one of the individual DARWINs. Over time we have seen that you cannot fix a value dependant on the dispersion of whichever type of operation, and the same for every one of the performance bars in the risk manager
This procedure systematically works against those operations that have a lot of positions in a month and all types of durations, closing positions excessively when in reality, the actual risk had not gone above the VaR of the strategy.
In the following example, we can see the problem:
The position highlighted in grey, with a 17 minute duration and D-Leverage of 10, in accordance with the old analysis, would have suffered an adjustment of 40.6% in the risk for the investors in the DARWIN created from this strategy.
The question is: had we not closed this position, would the VaR of the strategy have increased? The reply is most probably yes, but with a very reduced %. In the snapshot of the month tested, this strategy had a lot of positions (327 in total), that resulted in a VaR of 10.72%. That is to say that the contribution of this to the VaR, at most was 1/327 of the total (we can make this approximation because the other positions have a much longer duration).
Definitively, in the case that this position had been left open to the investors, the increase in VaR would not have been too significant, and, therefore, we can conclude that it wasn’t necessary to apply any adjustment.
Instead, it was closed because in the range of 15 to 30 minutes, the leverage was outside of the range of the maximum tolerated dispersion and was much higher than it’s historical median. The result in this case is that the trader closed the said position positively (+0.44%) , despite the profit of his DARWIN being seen to be reduced by 40.2%. It could be said therefore that in certain cases our risk manager was over-protecting the investor.
- New risk adjustment algorithm
In the moment of opening a position, the new risk manager analyses the maximum thresholds for each time interval that, if surpassed, would suppose a potential rise in risk to a VaR of 10% in an underlying strategy (to say, a 10% +1% of tolerance = 11% of VaR in the DARWIN).
The tolerance threshold of 10% that we have concluded upon is that which we see to be apt for whichever type of operation, and how they reflect the studies completed by our team of quants. Without wishing to enter into the details of the VaR calculation, so as not to be excessively complex, the most important conclusion is that this new adjustment works in an effective manner, limiting the risk of the DARWINS to 10%
What effects has this adjustment had over the DARWINS? The two most notable conclusions of this new adjustment are :
- On one hand, those strategies that were being excessively closed by the risk manager by overtrading, happened to behave “better”, in the sense that the DARWIN and the strategy is going to appear more similar).
- On the other hand, those strategies that operate with only a few trades every month, and with a short duration, will have smaller acceptable standard deviation in the dispersion of it’s tolerated leverage above it’s historical median (the less number of positions, each one has more weight in the VaR).
We hope that this series of articles has been useful to understand the changes in the Risk Manager that are in the offing. As always, at firstname.lastname@example.org we will be very happy to help you with whatever doubts you could have.