Interesting Concepts For Managing Our Automated Trading Systems
As our business develops and we build and deploy more strategies for our clients and our trading accounts, we wanted to start looking into total risk management for our automated trading strategies.
At current we let our strategies run regardless of the current performance. We rely on our testing to guide us through the highs and lows and ultimately have confidence in our systems to perform over the long term, under many different market conditions. But is there a better way to manage our money? Can we be more effective and place more weight on better performing strategies? How do we do this?
There are several methods which we are currently exploring to automate the process and I wanted to discuss these possible options in more detail in this post today. Both these concepts are very raw and are really our thoughts written out for all to see. Maybe these concepts will be the next big thing in portfolio management, I don’t know, however it’s something we are really getting excited about and the potential seems extremely powerful to the retail trader.
The first concept we have been working on is an equity curve standard deviation method. Basically in a nut shell we look at historical data and determine a moving average of the equity curve and then have a percentage standard deviation above and below this moving average. If the current equity curve is outside these standard deviations we will cease trading or change to simulation until performance is returned to the mean.
The idea behind this every trading strategy performs differently in varying market conditions and what we are effectively doing is trying to smooth out our equity curve to reduce our drawdown and ultimately allow us to make more money. This also leads into capital allocation as when a strategy is under performing and switched to simulation we can then allocate more money to the strategies performing the best.
This is a highly complex concept and there are many questions we need to ask and ultimately program into our model. Questions like what percentage deviations off the mean are going to be most productive? If a strategy is ceased due to being outside the prescribed levels when do we trade again and what should we look for to commence trading? Is it productive to stop trading if the current equity curve is outperforming the top deviation level? How much data do we look back to determine our equity curve, is shorter terms better than longer terms due to being more current? All these questions we are working though at current and we hope in the coming months to have something more solid.
The second concept we are working on ties into our equity curve standard deviation method somewhat and is based on capital allocation but on a real time basis. For this model we run a batch of strategies in simulation on as many markets as we like with varying time frames and settings, and then the strategy will monitor this simulation environment, every so many minutes and choose a certain number of strategies which are performing best.
Our latest strategy designs we are currently testing are higher frequency trading systems that take as many as hundreds of smaller trades per day which is perfect for this type of model.
For example let’s say we are trading 10 markets across 3 timeframes with 5 different settings per timeframe, this gives us 150 active strategies. Now we tell our strategy to monitor the simulated performance in real time by checking in every 10, 20 or 30 minutes and choosing the best 5 strategies. Those 5 strategies would have 20% of our capital allocated. Every 20 minutes the process will be repeated. If the strategies within the current batch are still the best performing they will stay, if not they will be replaced.
There is one question that sticks out to me if we were running this model which is, is back testing really necessary? If we are monitoring the market with an unlimited amount of settings across an unlimited amount of markets and time frames would back testing matter? When I look into this I don’t really know if it does. It is a completely dynamic real time management system which only looks at the real time results.
Let’s say for example we say back testing does matter. We use our equity curve management system along with our real time monitoring, then we could narrow down on the best performing strategy settings to reduce our sample size trading in simulation and then have our real time monitoring take over from there.
These two concepts as mentioned are very raw. Our research into the concepts is thin as there has been very little written on the subject that the average trader can get hold of. I believe these methods have been used in high frequency firms for some time, but we would never know. Their businesses are super secretive and we will never know how they manage their portfolios. For the average retail trader these concepts could hold the key to optimizing portfolio management of a basket of automated trading strategies.
Let us know what you think and be sure to look at our previous post The 5 most common reasons why people fail at automated trading and what you need for success.