Scott’s Journal

[ScottMcNab]

I was going over my paper trades on the weekend and comparing them to trades Amibroker would have taken when I noticed there were more trades in Amibroker backtest…turns out that on days where exited on close the backtest was using that equity to allow purchase of additional entry positions on the same day to make up the required number of stocks for max open positions…type of postdictive error I guess. After I corrected this by adding settlement delay the system performance tanked. Another lesson for benefits of paper trading but now need to work out a new system….again… I like the limit buy order with MRV so will see if I can’t come up with a simpler type of exit…maybe just sell on open for all exits to avoid making the same mistake again.

[Nick Radge]

hmmm…yes. And this is why we test in a simulation environment.

Are you using leverage? Can you tone down the leverage to facilitate both?

[ScottMcNab]

No leverage Nick. I think I over-complicated the exits. I have another MRV that I came up with last night that passed stress tests and I started paper trading overnight. 340 orders into API
3 fills
Amibroker backtest same as TWS trade record…so far so good…

[TrentRothall]

is that on the Rus3000 Scott? With 340 orders you would want the API to be working well!!

[ScottMcNab]

R1000…prob feel more comfortable in SPX ultimately but will see how goes when eventually get to trade live

[TrentRothall]

No worries, thanks

[ScottMcNab]

Giving myself headache re selection bias…..as long as the trade skipping test (1000-2000 runs of maybe 20%) returns an acceptable variance then system passes ? Or is the fact that system only taking 40-50% of available trades (as determined by increasing max positionsize to 200) still a worry regardless of it passing the trade skipping test ? I imagine the impact of selection bias would be a lot more in some systems where big winners are needed to make up for multiple small losses rather than a system with 60/40 w/l ratio and size of win same as size of loss ?

[SaidBitar]

For me I rum MCs with 25% trade skipping percentage and i look at the results, if the average is OK and the variance between the results is not huge then it passes.

the reason for this is the double random, the first one is in taking the trade and the other one is in ranking the trades that passed the first one

[Nick Radge]

Quote:

as long as the trade skipping test (1000-2000 runs of maybe 20%) returns an acceptable variance then system passes

This is correct.

Let’s assume two outcomes:

#1
You run 1000 MCS with 20% trade skipping, consider,

Avg CAGR 15%
Low CAGR 12%
Hi CAGR 18%

Regardless of the trades taken the system is profitable. The question you need to ask is if you get the worst trades, i.e. the 12% CAGR, will that be acceptable?

#2
You run 1000 MCS with 20% trade skipping, consider,

Avg CAGR 15%
Low CAGR -2%
Hi CAGR 32%

This one is not acceptable. It suggests that some selection of trades has a negative expectancy.

The same would go for other metrics.

Ideally you want the differences between high and low to be reasonably tight.

.

[ScottMcNab]

Thanks Nick and Said. I wasn’t understanding how CAGR could go up when reducing the number of trades by 25% but the higher value is, of course, in reference to the average CAGR returned for the test itself, not for the metrics for the system without the trade skipping….its a gift…the ability to create confusion in an otherwise straightforward situation :whistle:

[ScottMcNab]

Haven’t added to this for a while but been busy…

started again and now have a new MRV system that I am paper trading…..also working through the steps to get reg-t account….so needed pty ltd……then discussions with accountant and lawyer to find out type of entity it falls under for US tax law (passive NFFE in this case)…is a good post on aussie stock forums with info about it if anyone else going down this route…then trips to bank for account setup for pty ltd…which then need to submit for IB appn

Long term: looking to try and trade 2 different systems that both enter with buy limit….have the afldone…..one is mrv (buy limit entry pretty standard) but the other is a momentum system (so sounds stupid but entry with buy limit works)….so next I have to find out if can run two Batchtrader (to control buy limit orders overnight) through same TWS or if need two TWS accounts…..

[ScottMcNab]

Nice night for US MRV…4 new positions ( CBL, EXC, FX and TIF) which dropped for entry and then took off

[ScottMcNab]

Review of trades taken to date verses backtest shows 1 difference so far;

GIS
buy limit order placed for 24 May for 61.83 with Batchtrader
open 61.69
low 61.60
not filled/entered for some reason….all criteria in MC afl seem to have been met….may have just been one of those days where paper trading acct misses

Will continue to monitor

[Nick Radge]

Check the time and sales – it may have been a trade outside regular trading hours.

[ScottMcNab]

Thanks Nick, that seems to have been the case…nasdaq.com has the lowest price after open at 61.91…higher than Amibroker’s price of 61.69….61.69 is also the open price for GIS on nasdaq.com historical OHLC for the day…so learnt something new for day….thought OHLC was for regular trading hours

Also makes me wonder how many (most ?) of the open prices in Amibroker therefore reflect trades of outside regular hours and how this impacts backtest…. Should I consider changing TWS settings to allow trades outside of regular hours so that the Amibroker prices used for the backtest are same as ones obtain in paper/live trading ? Not much point having a backtest based on prices cant actually get ?? :unsure:

[Author]

Posts