I use Python's Beautiful Soup library to scrape the data from public sites like http://futures.tradingcharts.com/marketquotes/KC.html There is a delay, but if you're trading a longer time horizon that works OK.
Current situation: Trading 45 different instruments- £430k. Running EWMACs (4-64) + carry only. Backtested Sharpe of 0.85. My models include costs and spreads. I'm building a slippage model so I can try and trade faster (running EWMAC 2). My real question is- where do I go from here? I'm aiming for realised Sharpe > 1. I have some ideas about reformulating the system as an HMM, which seems promising in theory (to be determined if it will work), and building it out into a realtime system. If anyone has any experience of working with quants, there seem to be a lot available for hire on Upwork at prices I can afford- thoughts welcome.
Hi GAT do you know what type of strategy Rentech is using based on your experience? How are they able to beat everyone else?
I tried using HMM (hidden markov model) before. Just using price as a single factor will not give you an accurate model.
I don't remember all the details, but you can go try it. What are the states that you are gg to pick? What distribution of returns are you gg to assume? Price series are already not very predictive of future returns. Why do you think that using another model on the same data will lead to better results? Not trying to put you down but just wanna save you some time if you haven't thought abt this deeply.
Reasonable states would be, big drop, small drop, small gain, big gain (normalised as appropriate). This means you can answer the question - "how long might a trend last?"
I think in the book I say if you are running more than 16% annualised vol you need to check at least daily. GAT
My personal experience indicates you will get best marginal add on performance from (in descending order of added value): - completely different trading style (eg selling vol, equity long short, rates long short, stat arb) - diversifying across time - if you can model the costs properly or capture the slippage (eg market making style strategies) - doing the same thing on different asset classes - doing the same thing on different instruments - trying to do the same thing on the same instrument, but slightly better (eg adding more trend following rules) Mostly there is an effort penalty: it's a lot of work to setup a vol strategy alongside futures trend following; but not always (diversifying across more instruments is pretty easy, assuming you have the capital). Using an HMM strikes me as being in the final category, and hence an unlikely place to get massive added value out of sample. GAT