Data-science techniques are increasingly a necessity for investment managers. Machine-learning can identify historical relationships among a wide range of variables and how sectors perform – something impossible for humans to do alone.

HSBC has developed SAM – a Sector Allocation Model that uses a machine-learning algorithm to predict the likelihood of a sector outperforming over the next three months. Testing it with historical data, SAM has had an overall accuracy of 58 per cent since 2014 and using it in sector-allocation would have delivered returns 2.1 per cent above the benchmark on an annualised basis.

Determining which sectors to be over- or under-weight is an important part of the investment process. Machine learning can help develop a robust framework for making these decisions.

And recent developments in the explainability of machine-learning models’ predictions now makes these techniques accessible to the majority of professional investors, allowing them to enhance and augment forecasts with their own fundamental overlay.

SAM includes 24 variables that include several fundamental and technical indicators along with more macro-fundamental data derived from bond and commodity markets. We also include proprietary data, such as our analysis of corporate management sentiment on earnings calls, which can provide another potential source of alpha.

The success of any model is ultimately determined by performance of an investment strategy that uses its predictions. To put SAM to the test we backtested it with an investment strategy that went over or underweight a sector when the prediction for a sector outperforming exceeded predefined thresholds.

Such a strategy would have delivered annualised returns of 7.8 per cent since 2014, outperforming an equally-weighted sector benchmark by an annualised 2.1 per cent. Volatility was low, and so was the volatility of SAM’s predictions: on average its recommendations for each sector changed only every 1.5 months.

SAM’s 58 per cent overall accuracy suggests that most of the time it correctly predicts sectors’ relative performances, with no bias to forecasting either outperformance or underperformance. Accuracy is also relatively consistent across sectors, though telecoms achieved just 42 per cent, possibly because of the greater influence by regulatory and structural developments that the model cannot capture.

So how does the machine learning algorithm work? Well in simple terms, our predictive model is essentially a series of decision-trees where each tree learns from previous mistakes.

Models can be quite complex and a common criticism of machine-learning models is that they are ‘black boxes’ – hard to interpret. However, significant progress has been made in explaining these models. This is particularly important for investment professionals who want to understand the reasoning behind a model’s prediction.

SAM can do just that and its results are encouraging, suggesting our model can be a useful aid to sector allocation.

First published 29 June 2020.

Would you like to find out more? Click here to read the full report (you must be a subscriber to HSBC Global Research).

Disclosure appendix

More, collapsed
Will sustainability matter to consumers after COVID-19?
The pandemic catalysed ESG trends despite financial pressures
Join the conversation?

Join our Linkedin group to get an unparalleled view of macro and microeconomic events and trends from a bank that is a leader in both developed and emerging markets.