I've been doing some strat development on soft commodities and have some pretty good result on KC in backtesting. Anyone trading this currently ? What sort of slippage can I expect and any particular day to watch out for ? like the wednesday CL inventory report equivalent ?
To my knowledge, there's not a specific report or day of the month that you need to watch out for, like you do with API, EIA or USDA. The two largest producers and exporters that drive the bus are Brazil and Vietnam. Production estimates from their agric depts or trade associations can move the market as can estimates from influential analysts, but they're not on a fixed schedule. Changes in the weather and exchange stocks can certainly influence the market but these events typically unfold over days and weeks, not at 10A on Tuesday.
Yes, I trade it but it solely on a discretionary basis. Since about February, it's had some sharp moves and the volatility is likely to stick around for a while. Harvest is winding down in Brasil but the quantity remains a big unknown. Most people expect that the drought will have a negative effect on production for at least another year or two. If you're new to KC, keep an eye on your margin. For much of late Spring and early summer it was $8000+. With a recent *single* adjustment it's now about $5700. It can certainly zoom right back up if things get wild again. You'll probably want to keep an eye on the spread trade New York arabica v. London robusta even if you don't trade it. Same with NY v. London cocoa, which I believe is mostly a currency thing (dollars v. pounds) rather than a product differential.
Appreciate the advice, So I got this system for KC which is an intraday short only system, which was a result of my machine learning experiment trying to reverse engineer a collective 2 system intraday arabica. So far I got the short side pretty close, was unsuccessful deconstructing the long signal they use. It's still in prototype mode, am in the process of migrating the code for live trading. But probably can be traded discretionary too using custom indicators since its only about 1 -2 trades daily. Backtest is from 2007 - 2014, 2013-2014 is OOS Comissions assumption is 6$, Any idea whats a realistic slippage is like for 1 - 5 contracts ? Not quite comfortable yet since I'm rather new with the instrument, If possible we can exchange ideas more, maybe my data mining results can give you fresh insights for your discretionary trades. Attached the equity curve and annual results for 1 contract KC
Anyone would like to have this equity curve ! Are your sure there is no over fitting ? Or use of any future data ? Because it looks too good to be good. Did you use intra-day or end of day data ?
Positive no over fitting, only 3 parameters. No futures data used. Intra day data used. 32 Xeon cluster running non stop for 28 days with custom SVM and GP. using 512GB RAM I'll give you an example of what is overfit on my next post
I feel that we cant fully avoid overfitting because we select the best model out of a lot models. So even in pure randomness, we can find a valid model at 5% out of 20 models. I also look for models corresponding on what relationship I saw on the market. Some of this relationship can be random but we dont know yet. So I have always trading results below out of sample backtests even I use few parameters, simple models and paramets which make sens from a fundamental perspective. So, all my strategies are based on very simple models because I feel the signal is lost in so much noise. Did you experience any great success using SVM on real trading ?