Clone wiki

ML Classifier Gesture Recognition / how-can-i-use-quantized-classifier-to-make-money

How make money using quantized classifier

I have been asked the question "How can I use this to make money?" by several people. Hopefully this article will help answer the question. There are also some detailed tutorials available.

You could take the Quantized classifier engine for free and with reasonable work incorporate it into your existing trading process. This can range from a fairly small amount of work to have it help your human traders find the best entry points for symbols you already want to buy.

You could invest more work to integrate the Quantized classifier into an existing fully automated trading system.

If you use Quantized Classifier to assist your manual traders then you don't need to risk much up front, just have your traders record when the engine told them to buy and when they actually did buy and see if you would have made more profit following the systems advice. If you would have made more money following the systems advice then you will want to follow the predictions and may want to invest in more configurations to find more entry and exit points.

Strategy summary from a SPY prediction sample

performance summary example

The difficult expensive work is provided for free

I have provided the most difficult, expensive and highest risk portion of any stock trading system which is a prediction engine capable of providing signals that if followed can substantially improve the probability of profitable trades. This capability could be combined with a larger trading system to improve it's profitability or it can be used to help improve decision making about when to purchase symbols you want to own anyway.

Many large companies have invested tens of millions of dollars building prediction engines less capable that what I am giving away for free. My engine gives you the ability to start with a small investment and incrementally grow your investment as you gain confidence with the results. If you can't afford a million dollars to hire an expert and give them time to build a team to build a prediction engine never knowing if they will be successful then Quantized Classifier can give you a head start at substantially lower risk.

I have provided examples that include simple indicators and configurations that provide favorable prediction results. There is nearly infinite flexibility in producing new indicators that could improve the predictive capacity. The demonstration system uses fairly simple entry and exit logic that should be modified to reflect more sophisticated requirements.

Geeks, Physicists and Math wizzes history of success with prediction machines

My approach is statistical machine learning and signal processing techniques similar to those used by "renaissance technologies" and other notable technlogy based trading companies. The results from the Quantized Classifier engine demonstrate my ability to use statistics and machine learning to identify signal needed to improve market timing in what is essentially a random market. There is a reason that Renaissance hires algorithm gurus, physicists and and signal processing geeks like me rather than finance experts. When you study the history and evolution of automated trading you will find that many successful machine trading systems were built by people with algorithms, math and physics backgrounds rather than finance backgrounds. These people come with fundamental tools and mindsets well suited to finding signal in what most people perceive as random noise.

Incorporating into larger trading system

I have only provided the core prediction engine and examples of how to configure it to find profitable trades. You must wrap trading rules, Risk management rules, Portfolio balancing rules and plumbing to connect the prediction engine to your broker before it can be used in a fully automated trading system. This was an explicit choice.

I have built these layers before and found that most companies had their own preferences for those components anyway. The work required to hook my engine to your preferred trading infrastructure is considerably less complex and lower risk than producing a working prediction engine.

Next Steps

You are free to take my engine and incorporate it into your trading system. My suggestion is that you find a specific set of symbols you want to trade anyway and hire me to help you find indicators and configurations that will meet your trading system goals. I can build as much or as little plumbing as is needed to integrate the system with your existing infrastructure.

HFT Context

Some traders want a system that never holds a position over night. That places the system in the general arena with HFT that range from holding positions from a few seconds to a few hours. This can be a profitable area but it requires a large investment to buy high speed data pipes, high speed broker API access and a full time IT support staff to keep it running. I can not suggest focusing on a HFT system unless you have a lot of money along with a substantial apatite for initial risk.

I suggest that most people initially focus on hold times that range from a few hours to 10 days. I acknowledge there risks of a gap down type stop out but there are ways to manage this risk and it all works out in the averages if you are using good trading rules correctly.

The core prediction engine could be used all the way down to the tick level events but be aware that HFT is an area heavily invested by large companies who can afford to hire 50 guys just like me to tweak systems they have invested a lot of money in so it is heavily competitive. Some of these guys are spending a 1/4 million a month just for high speed connectivity with guaranteed execution times so it is difficult to compete with them without a substantial investment. In contrast in the 1 to 10 day holds seems to be a sweet spot where you are completing with less skilled day traders.

Thanks Joe Ellsworth CTO Bayes Analytic LLC CONTACT

I have published results from our sample configurations. It is the users responsibility to independently test and verify the system meets their needs. The software, articles and configurations are made available on a MIT license which limits my liability. Please remember that trading places money at risk. You should not risk money you can not afford to loose. Even great systems occasionally loose money and nobody can ensure that a major market changes will not negatively affect trading results.

Updated