Update on Trading Research

I came to the conclusion that none of the trading tools I developed are reliable. They can match market behavior for a while and make money, but then the relationship breaks down. In the long run there is no relationship between any of these indicators and returns.

I wrote a back-testing program for Turtle type trend-following models. This allows me to optimize the time periods to use to maximize profits. There is the potential for over-fitting and unstable relationships here too. The answer I think is to regularly re-optimize as the market changes. This re-optimization is easy to do. Given that a wide range of values is profitable in the exercise I did, I don't think there would be a sudden failure. We will see.

In the backtest there were 78 trades with a 48% win rate. But wins were on average 3.45 times larger than losses. The annualized Sharpe ratio is 2.2. Here there is a negative correlation between the initial risk taken (amount of money lost if the stop is triggered) and profits. That means it makes sense to bet bigger when the risk is lower:

I am now trading Bitcoin futures with the optimized algorithm (Long Bitcoin since yesterday). Next step is to see if there are other futures I could trade. Stock index futures aren't more profitable than just going long the index.

Doing More Trading Research

Seems like April is the time for me to think about trading. I developed a very simple mechanistic trend-following model for trading Bitcoin futures. I have placed orders in the market but they haven't triggered yet. Initially, I tried to be too clever, but quickly decided that just using mechanical rules will work better...

Now I am returning to thinking about more sophisticated models as well. Here are the results from a very simple mean reversion model – it goes short when stocks are strong and vice versa, with daily trades on the NASDAQ index:

This assumes perfect trades with no fees. It worked great until the end of 2008. Then it did nothing for three years and then started working again. But in the last five years it again went nowhere. Strangely, it looks a lot like the returns from trend-following over this period. Still, from 2005 to the present it returned 19% per year. I might be wrong, but I'm thinking that this is a benchmark for more sophisticated forecasts. If we can predict that we should trend follow rather than mean revert for a few of the worst days here, returns would improve a lot. But it has to be a very simple method that won't result in overfitting.

On the other hand, the NASDAQ 100 index itself returned 14.18% and go long with a stop if the market falls 1% or more intraday returned 19.85% a year.

Bitcoin

I thought it was time to buy Bitcoin when I heard that CBOE was dropping Bitcoin futures contracts. Supposedly, when everyone hates something, it's the time to buy it. I have also seen research, which argues for a big rise in Bitcoin. I set up a simple trading model and it confirmed that I should go long now. I got approval to trade CME Bitcoin futures on Monday. Looking at the stochastic oscillator on the chart above, I thought there would be a bit of a pullback and put in an order yesterday to buy at $4000 when the futures were trading at $4140 with a stop at $3845. I went to a meeting and when I came out  I saw Bitcoin was at $4700. So, trying to be too clever, I missed the boat. The problem is that the stop is still the same according to my trading model. One contract is 5 Bitcoins. I don't want to risk USD5000+ on this trade. So, I will need to wait, probably a couple of weeks before the stop will change and I can place a trade with acceptable risk. The alternative risk is that Bitcoin continues to go higher from here. But after yesterday, I really want to stick to the model and not try to be clever.

This is one of three more "macro" trades that I am thinking of doing. At the moment, I don't have the time to do the research to fix my shorter term trading model.