Discussing the article: "Codex Pipelines, from Python to MQL5, for Indicator Selection: A Multi-Quarter Analysis of the XLF ETF with Machine Learning"

[MetaQuotes]

Check out the new article: Codex Pipelines, from Python to MQL5, for Indicator Selection: A Multi-Quarter Analysis of the XLF ETF with Machine Learning.

We continue our look at how the selection of indicators can be pipelined when facing a ‘none-typical’ MetaTrader asset. MetaTrader 5 is primarily used to trade forex, and that is good given the liquidity on offer, however the case for trading outside of this ‘comfort-zone’, is growing bolder with not just the overnight rise of platforms like Robinhood, but also the relentless pursuit of an edge for most traders. We consider the XLF ETF for this article and also cap our revamped pipeline with a simple MLP.

The XLF ETF is a sector ETF by SPDR that represents the financial sector. Unlike the FXI we tested with the last time, it has had a predominant bullish trend since inception. Except for, the GFC period, this ETF has almost always rallied, and it thus gives us a contrasting testing ground to what we had the last time, where the FXI was languishing like a forex pair, about its 40 price handle for many years. Nonetheless, it does exhibit sharp regime transitions that are arguably driven by interest-rate policy, credit cycles, some liquidity shocks, as well as sector-specific rotation.

Its behavior over many quarters appears to alternate between smooth surges upwards and abrupt volatility spikes. By training and evaluating models on the XLF, we make the case for our pipeline, which is that indicators should remain consistent over different market sub-environments and that feature engineering should withstand noisy transitions. The ONNX model, our end product here, needs to survive real-time execution in MQL5 without relying on some forgiving conditions in benign markets.

Author: Stephen Njuki