Choosing an Inflation Model Horizon with Sensitivity Analysis

Summary

The document asks how to choose a long-run horizon when modeling euro-area inflation for a pension fund balance sheet. Because the inflation series is treated as non-stationary, the author proposes modeling its differences with a VAR and adjusting their long-run average so that the inflation level moves toward the European Central Bank’s 2% target. The proposed adjustment spreads the gap between the latest observation and the target across a selected number of years; the author has chosen 40 years without a specific rationale.

The response says that the horizon depends on the model’s purpose and accepts an arbitrary choice as a possible starting point. It recommends sensitivity analysis to see how alternative horizons affect model results. No empirical evidence, preferred horizon, or detailed sensitivity procedure is supplied, so the exchange does not establish that 40 years is appropriate. The practical takeaway is to treat the horizon as an assumption and assess its effect on the pension model rather than regard it as a universal constant.

Key ideas

• The author models differences in inflation with a VAR to address non-stationarity.
• The proposed adjustment aims to move the long-run inflation level toward the central bank target.
• The number of years used for this transition depends on the modeling use case.
• Sensitivity analysis can show how the selected horizon changes model outputs.

Tags

# How long is considered `long-term'?


# How long is considered `long-term'?












For a project I am doing I need to simulate the balance sheet of a pension fund. In order to do so I also need to simulate euro inflation. Since my inflation data is non-stationary, I model it using the differenced data. Since this means that the level data of inflation will take on extreme values in the future, I have the following idea:

Since I use a VAR model, I can influence the long-term average of the differenced inflation. I want to set this long-term average such that in the long-run it reaches the 2% ECB target. I do this by calculating the difference between the ECB target and my last observation, and divide it by the total number of timesteps until the 'long-term' is reached. Currently I use 40 years, but this is an arbitrary number.

Now my question is, what is a reasonable number of years to use as 'long-term' in this context?

## Answer by jaamor (score 0, accepted)

https://quant.stackexchange.com/a/42336

There's nothing wrong with picking an arbitrary number like 40 years for long term. It depends on your use case. I would do a sensitivity analysis to check how it impacts your model results, however.

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.