Choosing a Price Window to Evaluate Analyst Forecast Accuracy

Summary

The document raises a measurement question about evaluating analyst price targets with forecast horizons stated as a range, here twelve to eighteen months. One proposed method assigns the midpoint horizon to each report and compares the target with the closing price on that single date. The author worries that an individual close may be unusually noisy and could make the evaluation sensitive to an abnormal trading day.

A second proposal compares the target with an average price over a six-month window centered on the midpoint date. This would reduce reliance on one observation, but the document does not establish that it is standard practice or provide a definitive recommendation. It presents the choice as an issue in defining the forecast outcome: a single-date price measures accuracy at a specific horizon, while an averaging window measures accuracy against prices across a period. Any comparison should make that estimand explicit and consistently apply the chosen horizon and price convention.

Key ideas

• A forecast horizon expressed as a range requires a consistent rule for selecting an evaluation date.
• Using the midpoint of the stated horizon is one possible way to define that date.
• A single closing price can be affected by unusual trading on the selected day.
• A centered average window reduces dependence on one close but evaluates a different outcome.
• The document poses the methodological choice without citing an established standard or settling the question.

Tags

# What share price to use to calculate forecast accuracy?


# What share price to use to calculate forecast accuracy?












I've been reading about models like Thiel's-U but I can't figure out what I should use for the actual trading price.

The dilemma is this: the analyst reports I’m looking at are forecasting for 12-18 months into the future. I considered just adding 15 (average of 12 and 18) months to whenever the report was published and comparing their price target to the closing price on that single day (i.e., June 1, 2001 compared to September 1, 2002). But this seems to have a very small margin of error and is susceptible to abnormal trading days.

I also considered using a 6-month running average centered at 15 months after the report (i.e., the report from June 1, 2001 compared to the average price from June 1, 2002 to December 1, 2002). This seems like a much better option, but I wanted to know if there was already some established practice that I'm missing here.

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.