Comparing Projected and Actual Daily Returns

Summary

This document raises a model-validation question: how to compare two streams of net present value changes converted into daily return series, one projected and one realized. The author currently uses Excel’s LINEST regression and is unsure whether that is appropriate when both inputs are observed return series rather than a known explanatory variable and response variable.

The text does not recommend a method or report results. A regression can still be specified by treating one series as the predictor and the other as the response, but the choice should reflect the validation goal; it is not a symmetric comparison. Other useful checks could examine paired return differences, bias, error size, correlation, and performance across time or market conditions. These measures answer different questions and do not by themselves prove that a model is valid. The document provides no details on sample length, dependence, missing observations, or return construction, all of which affect interpretation.

Key ideas

• The comparison is between projected and realized daily changes in net present value.
• Regression requires assigning one series as predictor and the other as response, despite both being observed data.
• Paired differences and error measures can evaluate forecast bias and accuracy from complementary angles.
• Correlation alone does not establish that projected returns are accurate.
• The document poses the validation problem but provides no chosen method or empirical findings.

Tags

# Compare a timeseries of projected versus actual returns?


# Compare a timeseries of projected versus actual returns?












I am trying to validate the use of a model. I have two streams of NPV calculations, one is the actual return, the other is the projected return. Those streams are turned into deltas from one time period (1 day) to the next to create it's own time series (net from day-to-day)

The current model uses a regression approach to compare the delta results. The comparison is done in Excel with the LINEST formula. I am concerned with that approach because LINEST expects known X and known Y values. But if I am being given two streams, aren't those two Y values? The X values are still unknown and I don't think LINEST would apply in this scenario.

What other approaches can be used to compare two sources of return data?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.