Launched: Measuring Toxicity of Trading Strategies

23 April 2020
Javier Colón

We continue to add statistics of interest into our diagnostic kit. This time we’ve added the toxicity of a DARWIN’s and its investors’ flow.

We show the new statistics in the Capacity (Cp) study. At the moment they don’t affect the Cp score.


What is flow toxicity

Toxicity attempts to measure how easy it is for a liquidity provider to act as counterparty to a flow. We’ll see in a moment why it’s a very relevant parameter.

There are agents in the market who make the market by providing liquidity to the order book. They do this in the hope of obtaining a profit with the least possible risk. They take the liquidity from one side of the market and hope to get rid of it as soon as possible by demand on the other side. If they’re able to cross an order instantly, they earn the spread. But what happens if the market moves while they undo their position? Sometimes they’ll win and sometimes they’ll lose. On average, they’ll make a profit (or loss) depending on their ability to determine which orders to take.

The toxicity of a flow will then depend on:

  1. The volume of the order in relation to the volume that moves in the opposite direction in a period of time. The larger the volume of the order, the longer it will take for the provider to remove the risk as it won’t find enough demand in time on the other side.
  2. The direction the price takes in the moments after they receive the order. If the price moves in favor of the DARWIN, it means that the provider goes into a loss. If on average there is always a bias in favour of the DARWIN, they’ll consider the flow toxic.
  3. The type of supplier you have on the other side. In particular their risk appetite. That’s what will define how long they are willing to keep a trade and volume open on a given asset.

Why it matters to know the toxicity of the flow

Case 1

If a DARWIN is very toxic, we can assure that it won’t be able to manage more volume than estimated by the Capacity attribute. These DARWINs can cause problems of divergence even with low investment volume. Low meaning lower than the largest volume defined by the Capacity attribute. A DARWIN being toxic without having much volume can be an indicator that the provider is selling its signal elsewhere. Or that it’s an EA. We may not be seeing all the volume that the DARWIN is actually handling. It can also be an indicator of strategy exhaustion, because many other traders are attacking the same inefficiency, especially if it’s a strategy that seeks few pips on average per trade (less than 10 pips).

Case 2

A DARWIN with a high divergence but no toxic flow is suffering from divergence for a specific reason. This reason is not properly fractionating its operation. The market would most likely be able to absorb more volume. The trader needs to introduce the volume gradually into the market (at least 10 second of difference).


How we measure toxicity

From the perspective of the liquidity providers, the size of the order is not a problem, because they can manage it:

  • limiting the largest size they’re willing to put on the order book, and
  • defining the greatest exposure they wish to have in a given asset and flow (broker).

Each provider will define a limit according to their risk appetite. So, it’s a parameter that is in their control.

But, the direction of the market is something they cannot manage a priori. So they must protect themselves from taking orders from a flow that turns positive fast. They must also make sure not to keep orders from such flows longer than their risk tolerance allows. In summary, they must protect themselves from closing their trades at a loss.

Also, a very large order – considering the existing flow in that asset at that moment – will make the price move in favor of the order. The different providers will withdraw until there is interest on the other side of the market with which to protect themselves.

So, the toxicity of a DARWIN will depend on how the price moves in the moments immediately after it sends an order. Our statistics show this up to 5 minutes, to include many of the risk appetites of providers.

If the price of the underlying moves in favour of the strategy in a representative sample of orders, the flow will be toxic. Otherwise it will not. The sooner and the more the price and volume of the order moves in its favour, the more toxic it will be.


Darwinex Toxicity Tool

We wanted to make a tool that not only defines whether the strategy is toxic or not. We wanted it to help DARWIN managers to increase their capacity if necessary.

For each DARWIN, we take the last 200 opening and the last 200 closing orders. We present two graphs with each of the samples.

Return Chart

Here we measure the monthly return the DARWIN would have had if it had sent the order x seconds after it actually did.

To do this we take the prices of the underlying x seconds later using our tick-level price database. Then we calculate the trade’s return at that new price.

If the return is worse, the DARWIN, in that sample of 200 trades, did better than if it had opened x seconds later. In that case it would be toxic.

Showing the monthly return enables us to compare it with divergence which is also measured monthly. This in turn enables the manager to properly optimize capacity.

If a DARWIN has a divergence of -1%, but by fractionating it can get a -0.2%, you will know that you still have a fractionating margin. So, the tool is a great help in quantifying how much and how you can split the trades.

Many DARWINs have very different entry and exit toxicities. With this new tool you can also determine which orders to split and which not.

Bear in mind that the study includes 200 orders and does not exclude any. There may be orders that distort the chart in one direction or another.

Chart in pips

The second graph shows the difference in unit pips (difference in price divided by price and multiplied by 10000). The pips show the average difference in price the strategy would have gotten if it had placed the order x seconds later than it actually did.


We hope this new feature will be useful! Please give us as much feedback as you can to improve.