Of course it is.
Machine Learning Algo for Trading
- Thread starter stepseazy
- Start date
No, we do not sell strategies. I also use that old "It can't work, otherwise you would trade it yourself and not sell it" argument sometimes, but in fact this statement is nonsense in most cases. We're paid for developing a system from client's specifications. He still owns it. We could not even trade it unchanged for ourselves without violating the contract.
Last edited:
This is the key take away from this thread
Like
Setup a financial mode that makes sense *then* automate
IT / Quants don't become traders
Traders instruct IT / Quant to automate / optimise stuff that already makes sense
Different mind set.
When I make a trade, I am predicting for positive outcome. I have done much back testing on one minute ES and many other instruments, I got same results as you in the beginning, I didn't break down price enough, first I found best times to trade, volume controls markets and lack of volume is noise, and noise takes one type of pattern and can make it useless. Then I broke down Swings=pivot lows to pivot highs to pivot lows and found that there is an average and have to use variances above and below. SO you have trend and counter-trend. Where each signal falls within a Swing will give different probabilities of success, so I test for Swing ave on each hour, some hours are seldom traded as risk must expand on light volume and targets must get smaller. Eventually I broke down Swings into waves, so I ended up with mountain of stats and by defining signals and backtesting and optimizing smallest changes, get clearer idea of what can be 60% or better pattern, but where the money is in risk management of the patterns as this will even say whether pattern can be taken based on chart patterns to left of now.
I believe HFTs do so well as they know they are like a casino and have the odds on their sides, they know 95% don't have well back tested Trading Plan and often emotions plays on them, They know emotions, signals and where stops are kept from their back testing, lack of volume they see as opportunities, markets goes higher and higher then less volume, BAMM slam it on down after they feed the last to retail going on up.
oh please, I have been in this industry way too long to trust a firm sticks to contracts when a lot of money is involved. Not accusing your specific firm but nobody in their right mind would entrust someone with developing a strategy unless it is developed in-house and unless we talk retail accounts. Contracts unfortunately mean nothing when much money is involved. We have seen uncountable number of cases where contracts were violated or code stolen outright.
And it does not change anything about the fact that your former statements re input data being more important and more complex than the underlying algorithms is simply wrong.
Otherwise why do you think the world's top AI researchers are battling each other in high profile competitions where everyone starts off with the same identical data sets that are 100% clean and free of any errors? It is because they differentiate themselves by the underlying algorithms they develop and how they optimize their deep learning networks. This alone complete disproves your case.
Listen, I am not participating in this thread to deny you or disprove anyone for the fun of it, but when someone is factually wrong I think it should be pointed out and I proved sufficient arguments and proof that your line of argumentation is factually incorrect
https://www.kaggle.com/c/digit-recognizer
http://karpathy.github.io/2014/09/02/what-i-learned-from-competing-against-a-convnet-on-imagenet/
And it does not change anything about the fact that your former statements re input data being more important and more complex than the underlying algorithms is simply wrong.
Otherwise why do you think the world's top AI researchers are battling each other in high profile competitions where everyone starts off with the same identical data sets that are 100% clean and free of any errors? It is because they differentiate themselves by the underlying algorithms they develop and how they optimize their deep learning networks. This alone complete disproves your case.
Listen, I am not participating in this thread to deny you or disprove anyone for the fun of it, but when someone is factually wrong I think it should be pointed out and I proved sufficient arguments and proof that your line of argumentation is factually incorrect
https://www.kaggle.com/c/digit-recognizer
http://karpathy.github.io/2014/09/02/what-i-learned-from-competing-against-a-convnet-on-imagenet/
I was also planning to break down prices using moving average but now thank you so much for sharing your invaluable knowledge Handle123, surely, I will take the more sophisticated approach you described to me. I am currently testing GBPUSD but surely I will expand my research to other currency pairs as well as other assets in the future. Even with GBPUSD, I think it will take at least next 6 months to come up with convincing evidence but once I build sounding algo, then testing other assets will not take so long.
You told me you're a scalper but you sound more like a HFT, and to be a HFT, I am sure you have to have a concrete knowledge of statistics, implying that you're a researcher. Are you an academic?
Once again, thank you Handle123.
Last edited:
LOL, academic, far from it, have few college degrees-we all have quests of becoming something when young but then learn, WTF was I thinking moment. Am more of a hands on Mister Fix it individual, love to tinker till I accomplish, so love more of the research than the doing in trading. Love to have automation than having to sit there all day. Two years ago, by error, found two separate patterns within week of each. It is the closest thing to HFT I have tinkered, but both base themselves not on the small trader, don't really want to say more than that, but I don't do heavy size contracts, I want to remain like a ghost but automation does run 23 hours a day.
)Not to forget, the learning objective is also key. The return at the next bar is not always the best objective since it's strongly affected by noise. Better is a more distant prediction horizon, or something entirely different such as a Zigzag leg, or the presence or absence of a market inefficiency. The objective is even the most important key for supervised learning since it affects what the best inputs to the ML algo are. So, objective first, inputs is second, and third is selecting the best suited ML algo for processing them.
Last edited: