From various meetings and conversations, many of you already know that our Demography Sector has been hard at work for some time building a global equity forecast model. While we consider this Global Equity Forecast (GEF) model to be still in the R&D stage, we are confident enough about its preliminary results to begin to share some of them with you.
How the GEF Model Works. Currently, our model generates one-month-ahead forecasts on a strictly monthly basis. All forecasts are of MSCI indexes priced in USD. Each month, we close out our data collection at the end of U.S. market (4 pm New York time) on the 25th day of that month. Then, by the morning of the next day (the 26th), we publish our forecasts for equity returns starting at prices at the end of U.S. market that day and ending on the 26th of the following month.
- Rate of Return: For each MSCI index, the GEF uses the net total return above the U.S. 30-day constant-maturity Treasury rate. We call this the adjusted net total return (ANTR).
- Day of Month: When U.S. markets are closed on the 26th, the GEF defaults to the next open day.
- Indexes Included: The GEF projects ANTRs for 59 regional MSCI indexes. These cover all developed, emerging, and frontier markets, excepting only the following: Argentina, Bangladesh, Mauritius, Sri Lanka, Kuwait, Lebanon, Oman, Serbia, and UAE.
How the GEF Model Was Built. The GEF model is the product of over two years of intensive data collection and statistical research. We have gathered and tested roughly 200 monthly variables for each market going back to the 1990s. These variables (some of them highly proprietary) cover market, economic, demographic, and political trends. All told, the GEF model trains on over 3 million data points. Our statistical analysis is multi-stage and uses the most advanced machine-learning algorithms.
How Accurate is the GEF Model? Our accuracy is impressive. Our one-month-ahead forecast of each market’s ANTR averages over 0.5 RSQ for all markets, and better than 0.7 RSQ for the developed markets. For the average ANTR of the “Big 25” economies, our RSQ is 0.96 with a STDEV of 1.3 % points. For the ANTR of the total world (all 59 economies), our RSQ is 0.98 with a STDEV of 0.7 % points. To put this last number in perspective, keep in mind that the world ANTR’s historical absolute value averages 3.8 % points each month. Yet on any given month, the odds that our forecast will overshoot the next month’s actual return by more than 1.5 % points is about one in fifty.
GEF Model Track Record. In an appendix at the end of this note, we include four slides (two on the world average and two on the “Big 25” average) illustrating how well this model has worked since the inception of MSCI indexes in the late 1990s. They include our estimates for August. We have worked diligently to prevent overfitting the model to data, both by means of cross-validating while training the model and also by rigorous testing against out-of-sample data. In these slides, we show both in-sample and out-of-sample results. We have thus far never observed a deterioration in accuracy when moving out of sample.
Further Development. There is much development still in store for our GEF model. Please stay tuned. Most importantly, we plan to release forecasts on longer durations (all the way up to 3 years) and to run alternative independent variables—interest rates or credit spreads, for example. Through cluster analysis and other means, we plan to build explanatory tools able to shed more light on the dynamics behind our projections. Ultimately, we may be able to generate “now-cast” forecasts at any time of the month. These and other improvements are in store.
How the GEF Results Are Organized. Currently, we organize our month-ahead equity forecasts into two broad groups:
- Group Forecasts: We offer seven group forecasts, each of which generates an unweighted average of ANTRs. The Total World Index is the average of the ANTRs of all 59 markets. The Big 25 Index is the unweighted average of the ANTRs of the largest 25 economies (see note). And the remaining regional indexes divides up all 59 markets into regions. For each group, we generate a forecast of (a) average ANTR %, with STDEV of error term; and (b) intra-market dispersion, which tracks the STDEV of the ANTRs of all the member markets in the forecast month.
- Biggest Gainers and Losers: We also offer (out of the Big 25) our estimate of markets where we expect to see the largest % ANTR gains and the largest % ANTR declines.
Overall Picture for August 2018. Now let’s turn to the actual forecast.
SUMMARY: After undershooting a bit in July, the GEF continues to point to tranquil equity markets in August, with slightly negative % returns both for the World Index for nearly all subgroups. The absolute values of all group returns are within the GEF’s standard error. Intra-market dispersion is also expected to be lower than the historical average. Among the Big 25, the GEF identifies three most likely gainers and losers; GEF expects the largest movement from Turkey, to the downside.
APPENDIX: GEF MODEL TRACK RECORD
