The new Risk Manager 2.0 (Part 2)

7 March 2017
The Market Owl

As we explained in our last post, the launch of Darwinex reloaded will bring important changes in the functionality of our Risk Manager.

How you already know at this stage, the Risk Manager is one of the fundamental pillars of Darwinex, controlling the investor risk in an independent manner from the trader and maintaining the same risk for all the DARWINS.

The technical challenge behind the Darwinex algorithm is important: it must function for whichever type of trading system and under all market conditions… and all this in the shortest time possible so that the slippage from when the trader opens his position, until it is replicated for the investors, is minimized.

Apart from the announced changes in the last post, Darwinex reloaded will be accompanied with an important change in the new Risk Manager.

Reference period to calculate the VaR in the underlying strategy

The risk manager basically makes two adjustments in the leverage of the investors vs. the traders. Depending on how the risk in the account behaves in a proportional manner to the leverage of the underlying trades, the Risk Manager will apply the first adjustment in function of the VaR of the underlying strategy, with respect to the objective VaR of the investors.

Investor leverage  = Strategy leverage * VaR (target)/ VaR (strategy)

If the underlying strategy maintains a stable VaR over a long period, with just this first adjustment it would achieve the desired level of VaR for the investors.

However, the trader’s risk is usually unstable, thus it is necessary that we have a second adjustment, for the cases in which the trader over leverages outside of what he does habitually. We will explain this second adjustment in another post.

Centralising ourselves in the first formula (assuming the trader maintains his leverage in a stable manner), the leverage of the investors is very closely aligned with  the calculation of the VaR of the underlying strategy.

For it’s calculation, obviously, we have to build on the trader’s operational  history. The big dilemma is, how much trading history do we need as a reference? The answer is simple: it must take a period of time that would encompass the all the normal cycles of the strategy’s leverage. It sounds easy but the calculation is complex.

For example, we are going to analyse two different strategies.

Strategy 1

Strategy 1

The first strategy illustrates consistency in the leverage, the number of monthly operations and the length of these operations. That is to say, taking the history of the last month of it’s operations, we would obtain a representative shot of what we would hope this trader would do in the following month.

The leverage cycles are short, therefore that VaR value would always be very similar, (independently of the history that we have taken as a reference for this calculation).

Strategy 2

Strategy 2

Different to the previous strategy, the Strategy 2 causes more difficulties in choosing the reference timeframe for the VaR calculation because it has longer leverage cycles. The strategy, despite at first glance not appearing to be, does have discipline in his risk management, despite the leverage being very erratic. This strategy has long  periods of similar leverage over the two months. In this period of time, it can be seen that the trader apparently had a defined maximum risk (leverage) that didn’t pass the approximate level of 30.

For this second case, if we took values of less than two months for the calculation, the VaR value would vary a lot. On the other hand, if we took values of longer than two months, the VaR would be much more stable, always taking into account the peaks of leverage of the user. The VaR obtained would be a lot higher but, at the same time, representative of how the strategy behaves.

From 21 to 45 days of reference

Up until now, the calculation of the VaR has considered a period of 21 days in that a trader has had open positions. That is to say for example, that if the trader operates in one day out of every two, it would be considered to be a 42 day history. This period of time has been seen to be insufficiently representative and because of that we have decided to raise that reference period to 45 days (more than double that of which we have been using).

The principal effects of this change are:

1) VaR curves a bit higher than normal, but more stable for every strategy. This can result in a lower profitability in some DARWINS.

2) DARWIN curves a lot more similar to the underlying strategy, which is principally what we are looking for with this change.

From this situation, the doubt arises as to why we do not pick a larger  time period or why do we not adapt the selected time to the different trading styles.

The answer is that the greater the reference period, the greater the VaR would be and the less the sensitivity of the VaR curve would be to the behavioral changes of the underlying strategy.

Based on the different studies and analysis of our Quant team, 45 days is the optimum reference period to calculate the VaR without producing a significant loss of profitability in the DARWINS.

On the other hand, we have dismissed adapting the reference period of each type of trading style, because it would slow down the process of calculating the VaR (and as a consequence the function of the real time DARWIN Risk Manager). Besides, we understand that the criteria must be unique, so as not to avoid distinct results between different DARWINS.

A third post on the upcoming Risk Manager is underway! As ever, in we are available for whatever doubt that you may have.