3
$\begingroup$

let's say I am supervising a algorithmic trading project using machine learning. I don't have involvement in the technical side but am involved in the high level planning.

the style is likely momentum trading.

what are the resources that I can familiarize myself with the topics and what are the major algorithms already available to steer the project/conversations? thanks

$\endgroup$

closed as too broad by LocalVolatility, byouness, phdstudent, Helin, amdopt Jun 14 '18 at 17:40

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

5
$\begingroup$

'Machine learning' describes a very broad spectrum of algorithms. Just briefly here are a few conceptual areas;

  • Neural networks
  • Reinforcement learning
  • Genetic algorithms and genetic programming
  • Particle swarm optimisation (PSO)
  • Regression models
  • Optimisation routines
  • Markov models
  • Wavelet transforms and Fourier Transform and Spectral Analysis.
  • Clustering

You obviously can't go away and become an expert in those overnight but you can get a general feel for the types of problems they solve with a bit of wikipedia.

I suspect you are aware of this already but you can't just collect a massive amount of data, bung it in an holistic 'algorithm' and get results. The best results are often due to well placed conjectures on specific structures in data which precise algorithms seek. Many of the best results we see in 'ai' today are essentially hand-crafted algorithms designed for a specific task on specific data.

You will no doubt encounter two things in your project.

1) Backtesting - Make sure there is a testbed of data that an algorithm has access to and that it has no 'snooping' ability, i.e. it cannot access data that is otherwise not available at the time. A good example is dividend adjusted stock prices when the dividend is an 'unknown future value'. Testing on completely unseen data is the only way to really evaluate the performance otherwise you have biased tests and the results of tests will not match reality.

2) Combination of algorithms. Decoupled algorithms can be constructed and combined, where another algorithm has the responsibility of detecting the best combination. Again usually done via optimising some function like reward/risk of a combination of algos. This is where genetic algorithms / PSO can be employed. Heres a final link to an example of merging different algorithms which had a hybrid reward structure with a fun outcome https://sploid.gizmodo.com/microsofts-ai-just-shattered-the-ms-pac-man-high-score-1796091352 - better link https://blogs.microsoft.com/ai/divide-conquer-microsoft-researchers-used-ai-master-ms-pac-man/#sm.000jjtxcn14wufqgy2q23tdff39d0

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.