Forecast Data Adjustment

The Bayesian estimation methods included in the Estimator are statistical techniques that have been customized by NFA for financial situations. These particular Bayesian estimation methods are useful for including current forecasts, more generally held investor views, or any other quantifiable external information, in the Optimizer inputs. The content of this added information may differ significantly from estimates based on historical data.  

Different sources of information may vary in certainty or precision. For example, forecasted return uncertainty may be much greater for emerging markets than for long-term corporate bonds. The long-term corporate bonds forecast should thus have a stronger impact on the final estimates, since it is more precise. The forecast standard deviations provide a means of customizing the input estimation process to consider current information and levels of forecast certainty separately for each individual source of information.

Select how the input forecasts are used in the Bayes Forecasting drop down menu on the Forecasts Worksheet. If using forecasts and data in the analysis, which should be the normal state of affairs, the NFA Bayes option will be the final choice to create the result inputs for the Optimizer. The other options in this menu allow the user to examine the historical and forecast sides of the analysis separately.

• The Forecast option dictates that the estimation results rely on the forecast risk and return and forecast contrasts exclusively. A complete set of forecasts for all assets is necessary to produce a result for this selection. The forecasts may come from the single asset forecasts or from forecast contrasts, but there must be at least as many total as the number of assets, and a single asset forecast or a non-zero portfolio weight in a contrast for each asset. The results of this process are not generally recommended for optimization because this selection results in standard deviations and correlations that rely on the strength of your forecasts, rather than the risk characteristics of the assets. Moreover, the return estimates may be guesses for many assets or come from only partial information.

• The Historic option dictates that the estimation results rely on the historic return series exclusively. Prepare the historic results first.

• The NFA Bayes option combines the forecasts and historic return series. This is the recommended option. This option requires forecast inputs. If forecast inputs are not present, then this option will be the same as choosing the Historic option. Prepare the historic results first.

The Run Bayes button causes the Estimator to calculate the results according to the choice in the Bayes Forecasting drop down menu. When the Forecast or Historic options are selected, all Run Bayes does is copy your forecasts to the Forecast Data Table on the Results Worksheet; the results are unaffected by any Bayes estimation. The NFA Bayes computation combines the historic and forecasts using specially designed advanced statistical methods. In all the methods, the resulting risk and return estimates populate the Results Table on the Results Worksheet.  

Because the NFA Bayes method uses the correlations with the historic and forecast returns, the results are not always intuitive. The forecasts with high confidence influence all of the results through the correlations. If the correlations dictate that two assets' returns rise and fall together, or move oppositely, the Estimator has to balance that relationship with the historic and forecast returns. For example, if you forecast a high return for Asset A with high confidence and a low return for Asset B with low confidence, but the correlations indicate that these two assets run in tandem, the resulting return for Asset B is likely to be high despite your forecast.