This article was originally published in Arab Times
It’s that time of year again when market oracles debate what the coming year has in store. Economists and analysts are often criticised—with good reason—for inaccurate forecasting. In their defence, developing forecasts is difficult and riddled with challenges, especially in the GCC region, which is hostage to the volatile oil price. In addition many regional stock markets, are relatively new with little historical information available to perform rigorous statistical analysis. Economic data also comes with a time lag. This is further compounded as GCC stocks have only started receiving analyst’s coverage recently and therefore do not enjoy as much analyst’s attention as other emerging markets or developed markets. Following an earlier article in the Financial Times (FTfm supplement) Stephen Horan, CFA, and M.R. Raghu, president of CFA Kuwait discuss the role and challenges of forecasting in portfolio management.
What is the role of forecasting in portfolio management?
Formulating capital market expectations is the foundation of any sensible asset allocation strategy. We often equate forecasting with predicting future returns on asset classes, sectors, and individual securities. Importantly, however, portfolio management also requires managers to have a view on future volatility and correlations.
Moreover, we often think of forecasts in annual terms, especially at this time of year. Developing capital market expectations, however, is done in the context of the investor’s time horizon. In most situations, this requirement implies forecasting well beyond one year. Some of the challenges in making forecasts concern choosing relevant forecasting methods, properly interpreting historical data, overcoming behavioural biases, and managing the impact of forecast errors.
What are some common forecasting methods?
Ideal forecasting methods vary by asset class, but most forecasts have some connection with the past, to which analysts make a series of adjustments. A long historical sample period has the advantage of more statistical robustness but also runs the risk of introducing obsolete or irrelevant data. Statistical techniques allow analysts to place more weight on recent observations, de-emphasise or overemphasise extreme events, and capture the tendency of volatility to cluster over time.
Other forecasting techniques include adding appropriate risk premiums to the current risk-free rate, imputing the expected return implied by a discounted cash flow model or a portfolio optimisation model, and deriving financial market estimates from macroeconomic forecasts.
What are some data-related challenges in developing forecasts?
The timeliness and reliability of historical data are often questionable. For example, the International Monetary Fund sometimes reports macroeconomic data for developing countries with a lag of two years or more. Documents recently released by WikiLeaks reinforced the belief that some Chinese economic data may be “man-made.”
Hedge fund returns are notoriously plagued by survivorship bias, which is the inflation of average returns caused by the exclusion of failed funds from databases. Returns on real estate investments suffer from infrequent market valuations, which may not affect long-run return forecasts but which substantially reduces volatility estimates.
What are some of the behavioural challenges?
As human beings, we are vulnerable to several psychological traps when making predictions. We tend to think that the future will look like the recent past and naively extrapolate our most recent experiences into the future or over-emphasise experiences that have left a strong impression. We also tend to anchor our forecasts to our first impressions, over-emphasise low-probability events, and place greater weight on information that confirms rather than contradicts our pre-existing beliefs.
Most people are generally comfortable running with the herd. In making earnings projections, equity analysts (particularly those with little experience) often develop forecasts that are “in line” with other analysts’ forecasts. Finally, we tend to emphasise anecdotes and subjective personal experience over objective empirical data.
What are the implications of incorrect forecasts?
The impact of forecasting errors on portfolio construction depends on several factors. For example, if two asset classes have similar expected returns and variances, small changes in the inputs will lead to relatively large changes in optimisers’ outputs because the two asset classes are otherwise so similar. Optimizers will significantly over-allocate to those assets with either overestimated returns or underestimated variances.
Do these differences lead to large performance differences?
Interestingly, these types of misallocations have a relatively modest effect on the portfolio`s exposure to loss because the two asset classes are such close substitutes. Forecasting errors among dissimilar assets have an even smaller effect on asset allocation, which highlights the need to carefully define distinct assets in an asset allocation framework. Errors in estimating correlations have even less impact than errors in estimating expected returns.
What are the implications for portfolio management?
Although developing accurate forecasts is difficult work, we should not abandon the practice because it promotes insight and discipline. Furthermore, modest errors yield only modest differences in a portfolio optimisation context.
That said, analysts and portfolio managers are well-advised to avoid placing undue confidence in their forecasts and the models that use them. Most models assume that the expected returns, volatilities, correlations, and other inputs are known with certainty. In fact, they are only estimates of true values, and most traditional models do not fully incorporate this forecasting uncertainty. Therefore, be cautious and perhaps make conservative adjustments when interpreting a model’s outputs.