|
PureBytes Links
Trading Reference Links
|
(caveat: I ran this study quickly, there may be errors in the
conclusions and results.)
Here's a study along the lines that I mentioned. It trades
a continuous S&P futures contract from 1989 to 1996, every day.
However days with greater than 3% range are filtered out -
because they might bias the results, when we try to simulate
a certain degree of prediction accuracy. All gains, drawdowns,
are measured in percentages. Likewaise the daily slippage,
which I ballparked a $200 on today's (pre-S&P-split) terms
is is expressed as (200/500) / 1000, or approximately 0.04%.
The current margin rate is about 4%, but that wasn't used
in the calculations below. This system traded 1 contract only,
no compounding. Note: the buy/hold gain on the continuous
contract is a lot less than if I used S&P cash, because of
the effect of the futures "premium", which steadily decays
over time. Also, since this continuous contract is done with
an additive method, and not a percent change method, there may
be soem bias there, however for evaluation purposes this is good
enough. I plugged in a stop-loss limit of 1% of the current
futures price (9 points or so in current S&P terms, or $4500).
As you can see, as the accuracy goes down, the number of times
the stop is hit goes up.
Daily accuracy is simulated by calling a random number generator
that returns a float value on the interval 0 <= r < 1.0. If we're
simulating 90% accuracy and the random value r is less than or
equal to 0.9, we simulate a correct guess, otherwise we simulate
an incorrect guess. This process is repeated each day. With 1700
samples, we probably have enough samples to have high confidence
in our results being represntative.
slip: 0.04%
stop: 1.00%
filter: 3.00%
margin: 4.00%
trades from 891006 to 960712
buy and hold % gain: 47.9
num filtered out: 3
num trades: 1708
num up = 901
num down = 807
accuracy actual stop hits stop cost draw total
100 92 18 18.0 2.1 694.9
95 87 37 37.0 2.4 612.9
90 82 56 56.0 3.9 535.4
85 77 61 61.0 3.5 474.7
80 73 94 94.0 4.7 363.1
75 70 108 108.0 4.2 319.9
70 65 122 122.0 6.7 242.2
65 60 138 138.0 7.5 136.2
60 55 162 162.0 11.6 68.7
55 50 153 153.0 24.5 13.7
above:
accuracy - simulated accuracy
actual - actual accuracy, less than simulated due to stop
being hit. At 55% accuracy, the stop is hit roughly
10% of the time.
stop hits - number of times the stop was hit
stop cost - number of percent lost due to stop being hit. This
is total. In some cases the stop limited losses
and in other days, it limited gains. No differentiation
is made here.
draw - in percentage points from equity high to low point. To
put this in perspective current margin rates would be 4%
So, to trade at the 60% guess level, one would need about
three times the min. margin plus one, or 4 times the min.
margin to trade this system. That's 80K in today's terms.
Now, if the drawdown came along after a few years of trading,
you might be able to handle that. Probably, as a rule,
you'd need 100K (or 50K, post S&P split) to trade comfortably.
This gives you an effective leverage of about 5:1 trading
the futures.
total - total return on trading one contract, at the actual
accuracy shown, measured in percentage gain. Compare
this number to buy and hold. At the 60% accuracy level,
this strategy just beats out buy and hold (without taking
into account the traders time :))
Conclusion: to make this sort of daytrading method profitable,
the trader probably needs to be trading a method that is 65% to
70% accurate. If you've got a daytrading system with that level
of accuracy, please send it to me, right away.
Something to add, if compounding of contracts (trading increased
size as the equity grows), the returns skyrocket to very big numbers.
The conclusions remain the same however, something along the
lines of 60% to 65% accuracy are required to make the trading
system highly profitable/manageable.
--
--
| Gary Funck, Intrepid Technology, gary@xxxxxxxxxxxx, (650) 964-8135
|