Deploying ML to predict investment patterns in REIT funds
ML models can help to predict investment patterns in actively managed REIT funds and help analysts and investors to both stay ahead of likely trends and potentially harvest additional alpha from systematic or more focused approach to trading.
In this note we provide a summary of how deploying ML methods can help to predict likely investment patterns in global REIT funds. Being able to tilt probabilities in ones favour can have significant impact across the investment process such as impact decision-making of managing portfolios, creating alpha sources, risk management, improving timing of trades and ultimately improve alpha generation.
We use portfolio holdings data for global REIT funds with AUM in excess of ca USD 100M over the past 5 years and a range of stock level data and fundamental company metrics such as momentum, leverage, multiples, growth rates etc. (all adjusted for reporting lag) to study if this information can help to predict trading patterns during funds’ next reporting period.
In addition, we are less interested in nominal rebalancing trades and are more interested in what drives more meaningful changes in portfolio holdings, which we define at two levels; a fund’s position in the relevant stock has to change by at least 25% or by at least 50% respectively, i.e. the fund needs to increase or reduce it’s holding in the stock by at least that magnitude compared to the previous reporting period. Also, as investments, holding periods, portfolio rebalancing etc. differ between large-caps and small-caps, we split the data to address this based on on market-cap weights of stocks within the global REIT universe; with large-cap cluster being stocks with global weight of > 1% and small-cap <= 1%. We train separate ML models for the two clusters which aims to capture any structural differences in decision making and observed trading related to the two groups.
Finally, we compare the ML modelled fitted and predicted values with actual trading (changes in portfolio holdings) during the following reporting period. Fig.1 illustrates the output of this process trained on the more significant level of at least 50% change in holdings for the small-cap cluster. The axis represent the magnitude of change of a position, with the limits -1 representing complete exit of the position and 1 representing either entering a position or increasing it by 100% or more. The green dots represent successful predictions; two levels are illustrated, a) strict; predicted change in holdings of at least +/- 50% and actual outcomes of the same magnitude, and b) less strict; we relax the limit for actual outcomes to a change of at least 25%, i.e. predicted change in holdings of at least +/- 50% and actual outcomes of at least +/- 25% (faded green dots), while the white dots represent both nominal changes (smaller than +/- 25%) and incorrect outcomes.
The accuracy of the ML model can be measured as success ratio to correctly identify and predict future trading patters in terms of which stocks are likely to be rebalanced/traded in a significant way. For the strict group i.e. both predicted and actual trades with magnitude of at least 50% of the position that a fund is holding, the success ratio is 63%, while for the less strict group for which the magnitude of actual trades is at least 25% of the fund’s position, it increases to 68%.
In both cases the success ratio is high, which implies that it is possible to identify stocks in which meaningful trading is likely to take place and the direction of such trades, i.e. meaningful buy/sell patterns of individual stocks. This in turn implies that evaluating the universe on an ongoing basis can help to improve investment processes, timing, anticipate potential opportunities and ultimately improve alpha throughout the cycle in global REITs.
To discuss outputs for large-caps or to subscribe to this data set, please contact info@kaniaadvisors.com
About Kania Advisors
Kania Advisors is an independent research and advisory firm focused exclusively on institutional real assets allocations and investment programmes. We provide advice and solutions to improve outcomes in real assets investment programmes. We conduct detailed industry research and custom studies typically focused on quantitative analysis and provide insights which form a critical part of a client's decision process.