After many years of Central Banks propping up the markets, the long-awaited reversal of interest policy during 2022 brought into question the reputation of many a fund manager.
For active managers this was no more than the simple truth that contrary to popular belief they don’t always do well in volatile markets. What particularly stood out in 2022, though, was the realization that many passive manager’s offerings were also dead in the water. Who would have predicted that in 2022 in the UK, Vanguard’s lowest risk Life Strategy fund would lose 15.84%.
Now that Fixed Income assets are back in favour again, in this week’s brief article I discuss how at Algo-Chain we combine a Machine Learning model to process the latest macro-economic data with a set of technical analysis signals to determine which parts of the US Fixed Income spectrum one should allocate to, shown in the chart above.
Allocating to US Fixed Income ETFs & the challenge of too much choice
Perhaps not surprisingly after many rate hikes in the US, UK & Europe, an increasing number of institutional investors are allocating to Fixed Income assets. While there is value to be found, it’s not entirely clear which sub-asset classes offer the best opportunities. Focusing on USD Fixed Income ETFs, listed in the US, there are over 350 listings covering Treasuries, TIPS, Mortgage-Backed Securities (MBS), Municipal Bonds, High Yield, Floating Rate Bonds, etc with in excess of $1tr in AUM.
There’s almost too much choice and without any auxiliary tools, one will struggle. At Algo-Chain we have tracked the growth of the Fixed Income ETF market since the early days and have developed a range of models to address this challenge for example by harnessing the monthly data issued by the Organisation for Economic Co-operation and Development, OECD, for many of the key markets.
Based on the most recent market and macro-economic data, we have run our own internal Machine Learning model with a list of 17 different Fixed Income ETFs and show the most recent results.
Combining Machine Learning Models with Technical Trading Signals
Notwithstanding the fact that the OECD data is often lagging the current state of the economy, the basic idea is to employ a Machine Learning model to keep track of how each different type of Fixed Income ETF has historically performed across the various economic cycles and learn from that pattern.
Using the non-revised OECD’s Composite Leading Indicators for the US, which includes GDP data, along with business and consumer confidence data, it’s not unreasonable to expect an MBS ETF to perform differently to a TIPS ETF. What exactly that difference is, is not so easy to decode. As history evolves and provides more data for the models to analyse, the performance probabilities of the differing types of ETFs are updated accordingly.
Fixed Income ETF hit ratios provided Algo-Chain's Machine Learning model
Machine Learning models may be ingenious, but they have their limitations and unless one addresses this shortfall their usefulness is sub-optimal. To combat the slow nature of the Marco-Economic data, we combined this model with a z-scored Momentum signal, with the aim of increasing the hit ratio when forecasting future ETF returns.
In the table shown above one can see the hit ratios resulting from combining the two models. What we usually calculate, besides the hit ratio, is what we call the Hit Ratio Positive and Hit Ratio Negative. In this case, one calculates the hit ratio for all periods in which the signal was positive (negative). This makes sense for long-only investors, where one might be more interested in how good the signal is as a forecast, in case I am investing, than what I might have missed in returns because the signal told me not to invest, but subsequently the returns were positive.
The market data was analysed over a 10-year period and the first 5 years were used to train the machine learning model. The signals were calculated as of 30th January 2023.
Until next time.
Allan Lane