Hello Mr. Carver. I am new to trading consciously, I am learning a lot from your book Leveraged Trading and I would like to trade using your system in the book. However, according to recent news, there is talk that Turkey may impose a 40% tax on earnings from investments made abroad. Accordingly, warrants seem to be the only instruments in my country that allow me to invest in different markets as you suggest in your book (futures are almost exclusively offered for domestic markets). I would like to ask you: How can I apply the trading system from LT at least approximately to warrants? (I have been trading warrants for some time, but quite unconsciously) I would be happy to read any sources you recommend me to read for this, as I am seriously interested in systematic trading and your book has inspired me a lot. Or would you recommend me to trade with a broker like IBKR after all? Or anything else I can't think of? It is a pleasure to read your book and thank you very much again for all you have taught me.
Fully automated futures trading
- Thread starter globalarbtrader
- Start date
Hey Rob,
I was doing some digging into various rules and stumbled upon a weird thing. The raw and smoothed forecasts for breakout have interesting distributions.
That's for S&P500, 24 years of historical data. I double checked the data and code for the rule, even used your implementation from pysystemtrade, and I get the same thing.
A lot of other instruments have the same fat tailed distributions, though perhaps tails are a bit more even.
You mention in your latest book, Strategy 24, that it's tough to work with weird forecast distributions, especially bimodal. Yet, breakout is profitable both in live and backtest for me, how do we explain that?
Average forecasts do come out just above 10, which is adjusted with a forecast scalar so mean is around 10 (depending on the instrument), so that works too.
I got to this point by testing some new rules, got a pretty clear bimodal distribution, then went back to see what my existing rules look like.
I was doing some digging into various rules and stumbled upon a weird thing. The raw and smoothed forecasts for breakout have interesting distributions.
That's for S&P500, 24 years of historical data. I double checked the data and code for the rule, even used your implementation from pysystemtrade, and I get the same thing.
A lot of other instruments have the same fat tailed distributions, though perhaps tails are a bit more even.
You mention in your latest book, Strategy 24, that it's tough to work with weird forecast distributions, especially bimodal. Yet, breakout is profitable both in live and backtest for me, how do we explain that?
Average forecasts do come out just above 10, which is adjusted with a forecast scalar so mean is around 10 (depending on the instrument), so that works too.
I got to this point by testing some new rules, got a pretty clear bimodal distribution, then went back to see what my existing rules look like.
Thinking about this some more, why is a bimodal distribution an issue? If we had a bimodal distribution, with peaks at, say, -5 and 5, would that not be ok for our use case here? We could use 2 as the forecast scalar to get the average absolute forecast of ~10.
I'm sure I'm missing something here.
I'm sure I'm missing something here.
Breakout is a particularly weird forecast, because it's bounded - the price can never be higher than the recent maximum or lower than the recent minimum, only equal to them. Hence you would expect some 'squashing up' at the extremes. The smoothing brings things in slightly, but not much. Obviously for the S&P 500 we spend more time at the positive end because it's gone up more often that not.
Breakout is unusual in this respect. It's weirdness is a result of me taking a very binary concept (eithier the market has broken out or a range, or it hasnt') and shoehorning it into a contionous framework. Most of my trading rules have a reasonably vaguely Gaussian distribution, partly because they can't be bounded. All the EWMAC based rules do for example. But not all. Slower rules will generally have lumpy distributions, the slowest I trade is 'mrinasset1000' and that is very lumpy; whilst fast EMWAC is very close to Gaussian. And carry distributions vary quite a lot depending on the instrument, some are Gaussian, some have very fat tails, some are closer to lumpy uniform.
"it's tough to work with weird forecast distributions, especially bimodal. Yet, breakout is profitable both in live and backtest for me, how do we explain that?"
It's tough doesn't mean it's unprofitable! A binary version of almost all of my rules will still be profitable - yet that has a perfectly bimodal distribution. It's just easier to work with 'nicer' distributions, for many reasons; just generally the maths of portfolio construction works best if you assuming your underlying is Gaussian and linearly correlated (and if your forecasts are Guassian, and so is your price, and you can predict volatility with 100% certainty, then your final portfolio returns will also be Gaussian- none of those things is true, but the closer to the truth the better).
Rob
Hi all,
I have a general question about purchasing forecasts. I am interested in purchasing some forecasts for various futures markets and I am wondering if there is anyone selling high quality forecasts out there. I am primarily looking for short term forecasts, for the next day and week.
As of yet I haven't been able to find anyone doing this but I would imagine there are people out there participating in this kind of business.
Any leads would be appreciated.
Thank you,
rcgalbo
I have a general question about purchasing forecasts. I am interested in purchasing some forecasts for various futures markets and I am wondering if there is anyone selling high quality forecasts out there. I am primarily looking for short term forecasts, for the next day and week.
As of yet I haven't been able to find anyone doing this but I would imagine there are people out there participating in this kind of business.
Any leads would be appreciated.
Thank you,
rcgalbo
There is a side business for
someone here… How much are you willing to pay?
Hi Rob,
I have a question on pysystemtrade and pairs trading. I would like to trade pairs within the same asset class using z-scores and percentile rank as a signal. Is there a way to use the same rule structure to define the pairs and call the rule using the same logic you've devised for all other ones? The problem I have is when to form those pairs as what should be happening is that we first create the combinations within the various asset classes and then feed relevant instrument information. Right now the code runs for all instrument in the list and I'm almost thinking that I should write something completely separate from it which first works out the pairs, the applies the rule.
Thanks
I have a question on pysystemtrade and pairs trading. I would like to trade pairs within the same asset class using z-scores and percentile rank as a signal. Is there a way to use the same rule structure to define the pairs and call the rule using the same logic you've devised for all other ones? The problem I have is when to form those pairs as what should be happening is that we first create the combinations within the various asset classes and then feed relevant instrument information. Right now the code runs for all instrument in the list and I'm almost thinking that I should write something completely separate from it which first works out the pairs, the applies the rule.
Thanks