Backtesting 1987

You would want to make sure that your trading model put you in the right direction in the 1987 crash (which I am old enough to remember very well), wouldn't you? So, I backtested the model for 1986-87. The main model would be short going into the crash. But a more primitive model I am using in conjunction with the main model would switch to long on the Friday before the crash. That day the market went down 5%, so it would have already been a bad idea on the Friday. Recently, this secondary model has been doing well and I have combined its signals with my main model. So, we need some new rules about how and why to combine them. In this chart you can see that the buy signal would have come with the price already outside the +/- 2 standard deviations envelope (S&P 500 index):

These are "Bollinger Bands", though I use a 34 day moving average instead of Bollinger's 20 day MA. So, the new rule is not to take that signal when the price is outside the Bollinger Bands and the width of the Bollinger Bands is increasing. That wouldn't change much recently (NASDAQ 100 index):

The secondary model gave some very good buy signals just as price hit the Bollinger Bands in early February and late March. In these cases the price was not outside the Bollinger Bands or they weren't expanding.

The model is short for Monday.

How Big Should the Trading Program Be?

At the moment I am still in the experimental phase of the trading program. A 1 contract S&P or NASDAQ position either adds or subtracts about 0.1 beta to the portfolio. So if the beta of our portfolio to the market was 1.0, trading modifies this to 0.9 when short to 1.1 when long. My goal is to be able to hedge our portfolio against a market crash. That means we need to subtract up to 1 full beta from the portfolio. On the long side we then would double exposure. This means that the trading program needs eventually to be 10 times the size it is now. Using 3 times leverage on the cash in the trading account that implies allocating 25% of assets to trading. My existing allocation has 25% of assets allocated to managed futures. This total could be allocated between my own trading and "outside managers" such as Winton and meet this goal.

Why 3 times leverage? Simulation shows that about a 12% drawdown is possible. Remember that we use stops and or hedging to limit possible daily losses. So this drawdown means a string of large losses. With 3 times leverage that would wipe out 1/3 of the trading account. More than that and it will reduce the earning potential of the account too much going forward, I think. And be way too scary.

Formal Rules for Stops

I have decided on formal rules for setting daily stop losses. It is based on the pivot-point method. The pivot point is the average of the high, low, and close for the previous day. When short the stop loss is set at the second resistance level - the pivot point plus the previous day's high-low range - and when long it is set at the second support level - the pivot point minus the previous day's high-low range. If this results in a stop that is less than 1% from the opening price, I instead set a 1% stop. These stops increase the Sharpe ratio of the model though they slightly decrease returns. The chart below shows the last month of daily pivot points and first and second resistance levels:

The model got stopped out on 26 April when short - losing 1.42% that day. The market closed up 2.08%. So that saved 0.66% of losses The model also got stopped out on 3 May when long losing 1%. The market closed only down 0.02%. So that increased loss by 0.98%. That shows you why this reduces returns...

These numbers don't quite match what you can see on the chart as the chart shows the 24/5 futures market and the model is based on the NASDAQ 100 index. I am thinking of switching the model to use actual futures prices. Will need to pay for the data, I think.