Australian (ASX) Stock Market Forum

198% CAR - too good to be true?

Jump to new
gooloo said:
If I set the volume filter to C*V > 250000 the results are -

start - 01/01/1995
end - 01/06/2008
total trades - 13736
CAR - 30%
Max Sys DD - 16%
% winners - 31%
Click to expand...

ok, given that a slight adjustment in the volume threshhold has significantly reduced your CAR it would be worth paper trading your current system for a while and check the market depth to see if you are able to trade the low volume stocks without "moving the market" so to speak.

edit: I notice you're trading at the open, so check to see that opening volume is sufficient for you to be able to enter without affecting the opening price.

Also worthwhile is to perform a walk-forward test year by year(ie, 99-00, 00-01, etc.) just using a dummy variable

ie: dummy=optimize("dummy",1,1,1,1);

so that the walk forward test is still performed but you'll get an idea of what market conditions are more suitable for your system. If you find that you have negative returns during "bearish" periods you may wish to include some sort of index or composite based filter then develop a short system that is "switched on" during the time your current long system is "switched off".
 
the problem with paper trading (with eod data anyway) is that it's not going to show me slippage. It's also not going to show me if I'm moving the market as I'm not actually in the market! I guess sooner or later I need to consider using tick data.

thanks heaps for the tip on the dummy variable for walk fwd testing!!! so much quicker than the way i was doing it (running individual backtests year by year) :eek:

the million dollar question is.... has anyone worked out a "golden rule" for position size versus the average Close*Volume?
i.e. if I'm trading $10,000 worth of shares I can probably get by on a Close*Volume of around 50,000. but if I'm trading $100,000 worth of shares then there's no way I can do this on a stock with a Close*Volume of 50,000 without moving the market.

I'm toying with some ideas in my head around how to meaningfully back test this to come up with a magic ratio. interested to hear others ideas on this "magic ratio".
 
I limit purchase volume to either 5% or 10% of the average daily volume, and a similar percentage of the actual entry day's volume (although in real trading you won't know that until the end of the day).

If your position size is not based on equity, you can do this in the main code, otherwise you need to use the custom backtester.

GP
 
gooloo,

The standard error (which is derived from the variances of the equity line) is ridiculously high and suggests you have perhaps a couple of trades with abnormally high profits.

In the backtester results window sort the trades by profit and then examine in detail the trade with the highest profit to satisfy yourself that trade was actually able to be done at the entry and exit prices shown.
 
gooloo said:
thanks everyone for your thoughts - i'll try and answer all questions below....

by CAR i'm referring to cumulative average return.
the universe is the entire asx
i have a volume filter set - the 20 day volume average must be greater than 50,000

trade delays are set to enter and exit on next days open, stops are intraday
with no brokerage the results are -

start - 01/01/1995
end - 01/06/2008
total trades - 16718
CAR - 198%
Max Sys DD - 11.3%
% winners - 29%

with brokerage at $15 per trade the results are -

start - 01/01/1995
end - 01/06/2008
total trades - 16703
CAR - 185%
Max Sys DD - 11.3%
% winners - 29%

with trades set to buy on next days HIGH (rather than on open) and sell on next days LOW (rather than on open) i hit 100% drawdown.

for interests sake.... some other stats from amibroker's backtest report -

Recovery Factor 17.15
CAR/MaxDD 17.58
RAR/MaxDD 21.22
Profit Factor 2.27
Payoff Ratio 5.56
Standard Error 23865153571.62
Risk-Reward Ratio 0.14
Ulcer Index 1.94
Ulcer Performance Index 99.72
Sharpe Ratio of trades 2.17
K-Ratio 0.0092

the standard error seems a big number - i'm struggling to find a definition as to how this is calculated and how to interpret it.
Click to expand...

Hi Gooloo --

Several things stand out as problematical --

In AmiBroker, CAR is Compound Annual rate of Return.

Max system drawdown of 11% is outstandingly good for a CAR of 198%.

An average of five trades per day over 13 years -- how many symbols in the watchlist?

The K-ratio is too low.

29% winners is pretty low for a lot of people.

1995 to the present suggests that all of this is in-sample. Is that true? If so, it is probably meaningless. Run some walk-forward tests and see what the out-of-sample results look like.


Thanks,
Howard
 
Top