Cryptocurrency market analysis and algorithmic trading

It's possible to get a program to trade automatically. Finding out whether its method has real predictive power is much harder.

A method can look excellent on historical data and fail completely on new data. Trading costs, risks, market conditions and chance also affect the results.

I'm particularly interested in how to distinguish methods that genuinely work from those that only appear to work because they've been tried enough times.

Detailed image description

The chart is headed ICX and its original caption reads Table 33, Icon value. The vertical axis runs from 0 to 3.5. Neither the vertical axis unit nor the horizontal axis time period is labelled, so no assumptions are made about them. The curve has two high peaks near the start, a smaller rise in the middle and a low level at the end. It illustrates price variation, not the returns of a trading method or findings from the new research.

I'm about to begin a new, broader study of how well technical analysis works for established and smaller cryptocurrencies.

I'm planning sustained research on the subject, with the aim of publishing peer-reviewed scientific articles and possibly working towards an article-based doctoral dissertation later on.

The plan is to account for trading costs, compare results with simple benchmark strategies and investigate how well the findings hold up under different testing setups.

Finance Research Letters is one possible publication venue I've considered. No venue has been decided yet.

In the longer term, I also want to investigate machine learning and AI for market prediction. The interesting question isn't just which model predicts best, but whether its predictions have practical financial value.