Hi ETers, I'd like to review techniques relating to quantitative trading. 1. Stochastic Calculus I understand it's uses for assets and options pricing, but have anybody successfully use it for alpha generation in futures or fx trading ? I came across this http://stats.lse.ac.uk/kalogeropoulos/LD_1103.pdf After trying to implement the model outlined in the above paper, My resulting model ends up generating signals very similar to an EMA cross. Can the wizards in ET show me where I went wrong ? Also there is a relatively new paper by Steve Shreve. http://www.math.cmu.edu/users/shreve/FuturesTrading.pdf Too many maths, my head hurts, what does the above paper means ? But Steve says that futures is an arithmetic brownian motion, contrary to the DB quant in the first paper that assumes a gemoetric brownian motion. Which one do you guys use for your models ? Anybody using the discrete binomial or trinomial model ? 2. Genetic Programming I've had some luck using this approach, any tips on setup you guys use for distributed GP ? I'm using a java base JNI approach. 3. Neural Network Having used NN for various pattern recognition work in the past I fail to find its use for financial time series. If any NN users here can explain how they use it to trade 4. SVM Got some intresting results using SVM with a linear kernel. Whats your experience ? any suggestions on kernel type and c value ? 5. Clustering For me i'm still using normal k-means, what are others popular clustering method for financial time series ? 6. Bayesian Statistics. Anybody using this in their model ? 7. Backtesting Platform for tick data I'm using Multicharts.Net with bar magnifier, what about other ETers ? I hope the masters of the universe in ET can chime in and start the ball rolling here
You might wanna check out forums like nuclearphynance and willmott.com which are geared more for quants and wannabe quants, too
Well I am trading Live my very first system. Pretty good so far. So any stochastic calculus users or NN users here ?
stoch calcy kind of stuff only for my day job, not for my PA. the one system I have in my PA uses some different ideas (statistical). less of a fan of NN than I used to be. I feel its more time and effort than its worth in the end. but then, who knows, it might all be a reflection of my incompetence.
ugh, just lost a longer reply. anyways a disclaimer, my direct quant-ier experience is more related to derivatives pricing and risk mgmt on the buy side. still a work in progress in terms of standalone algo/system building. had a quick skim of the paper and I think that approaches like that are interesting and have potential but the implementation was not that interesting. In other words, I think techniques like that are just powerful tool sets BUT if applied directly on close to close data like the guy has, it does not really outperform simpler trend systems after costs, slippage etc. Price data leaves other footprints (I think the more experienced price action traders have specific terminology for it that I am not familiar with) which are more useful to model directly. IMO it is this dimension reduction or dimension translation step that is the harder (and more profitable) key step.
Thanks for the input, My suspicions is confirmed then, I'll stick to linear Machine Learning for now.
No, they are all ruined and have quit trading I hope you get the message. If not, you can try yourself.
Hahaha. It's just that I feel the math toolbox that I'm using for trading is too simple. Got an urge to abstract away and build nice models.