|
PureBytes Links
Trading Reference Links
|
Let's call it research. For starters there has to be a framework. For
example you might want to define a test condition, like "take profits $2000
and losses at $1000", and then look at the success rate of a given entry.
There are a number of reasons why a backtested system could fail. It's
important to test on as much data as possible. If the system is based on
REAL and REPEATABLE market patterns then it should work across a variety of
markets.
There should be a mechanism you can explain in terms of cause and effect.
This mostly rules out systems based on traditional "indicators". Basically
there's no reason why a market should turn (or continue) at any particular
value of a moving average, stochastic, or whatever. These kinds of systems
will work sometimes, and then crash and burn at other times. Any
periodicity will quickly dissappear.
There could be technical problems with the system, such as uncontrolled
risks. There should be some mechanism to protect against unforseen
conditions. Position size should be determined by volatility and potential
risk. The stops could be just a little too tight when things get wild.
Backtesting is really an exercise in statistics and probabilities. Best
results may be obtained by using an assortment of systems based on
different principles. The results are more valid given a larger number of
trades. For example, there's an extremely high chance the sun will come up
tomorrow, since it has so many times before. Note: There are those who
doubt the validity of this line of reasoning.
And don't forget the obvious questions should be answered, like "how well
does it do in down markets" (assuming a long trade). The numbers should be
realistic. There should be significant profit per trade compared to costs.
There could also be human-related problems in the implementation. I'm sure
we're all familiar with those!
But the most likely reason I can see for a system to fail is that it was
developed with TradeStation and that it uses the INTERNAL STOPS.
|