CunAruba Stock Forecasting System
Q: Let me begin by saying, I don't know a whole lot about trading stocks - but I do know a lot about forecasting time-series data, like sound waves, or stock charts. I write and use "self-learning systems". I use a number of techniques including genetic algorithms, neural networks, and chicken entrails.
A:
To me, forecasting stocks is all a numbers game, and I simply try to skew
the odds. I know nothing about balance sheets, p/e ratios, dividends, or
any of that stuff. I couldn't even tell you the names of the companies
involved, unless it was something obvious like "IBM". All I care about is
whether I can get in to and out of profitable t
profitable trades.
I have devised a "mechanical" trading system that seems to work very well.
I have tested it on "virgin data" up through 1996, and will be able to
test it on 1997 data this coming weekend - which I am really looking
forward to. At that time, I will also be updating the system daily - and
will be able to produce current forecasts.
Before I do that, and potentially make a total ass out of myself - I want
to be sure my methods of simulating trades are valid, which is why I am
going to share them with you. I need feedback. For example, are my
simulated trades unrealistic for any reason - and how can I fix that?
To recap how the CunAruba system works (this has changed since my last
post): Every 10 business days (2 weeks), you buy 2 stocks. These stocks
will be priced between $5-$20. If it is Wednesday night, and I tell you to
buy ABC and DEF, you buy them Thursday. For simulation purposes, I would
calculate your purchase price as the average of Wednesday and Thursday's
Closing prices. If ABC closed at $15 on Wednesday and $17 on Thursday, I
would assume you got in at $16 sometime during Thursday. Once you own the
stock, you keep it for 10 days, or sell either one that reaches +/- 30% of
Wednesday's price. Using +/- 30% stops doesn't really help, but a lot of
people wont trade without a stop loss. If the stock CLOSES at +/- 30% you
sell the next day, don't even bother watching intraday price swings. Just
follow the system. On the 10th day, you sell both stocks. I assume the
sale price as the average of the close between the 9th and 10th days. I
also calculate the commission rate as 1%.
Instead of using the average of today and tomorrow for your
buy and sell prices, you might consider calculating profits based on
the worst case (buy at highest price of the day, sell at lowest
price of the day) and best case (buy at lowest price of the day, sell
at highest price of the day). Alternately, you might consider the
midpoint between the high and low of the day as the buy/sell price, or
perhaps the midpoint of the high and low in the first three hours of
the trading day (assuming you will enter your trades in the morning),
or perhaps the midpoint of the high and low in the last three hours
of the trading day (assuming you will enter your trades in the
afternoon).
And then present both "best case" and "worst case" annualized returns?
I could do that, but I was hoping to give one realistic number, if
possible.
I might do this - I had considered dividing how far a stock went on the
second day by the number of hours the market was open, and then using that
to simulate a morning trade. I could use the 3/8's of the stocks move to
simulate a trade 3/8's of the way through the day. I do plan on treating
them as "morning trades".