Handling Missing Fund Returns in Mutual Fund Rating Models

Summary

The document considers how to build a machine-learning model to predict mutual fund ratings when funds have different operating histories. The dataset includes returns over several lookback periods, while some newer funds lack the longer-period observations. The question is whether those unavailable returns should be imputed or replaced with zero.

The answer rejects fabricating historical returns as financially unsound and stresses that model requirements do not justify distorting the data. It offers two alternatives: restrict the analysis to a common period no longer than the shortest available history, or retain only funds with sufficient observations for the desired lookback periods. These choices trade longer histories and a broader sample against consistent feature availability and sample coverage. The document gives no comparison of modeling methods or empirical results, so it does not establish which alternative will perform better. In practice, the choice should reflect the prediction target and the meaning of each return feature; missing history is not equivalent to a zero return.

Key ideas

• A fund cannot have realized returns for periods before its inception.
• Imputing unavailable long-term returns can create misleading financial inputs.
• One option is to use a common lookback period supported by the available histories.
• Another option is to exclude funds that lack the required return data.
• The document reports no empirical comparison of these approaches.

Tags

# Mutual fund rating predictions


# Mutual fund rating predictions












I am working on a dataset with aim to predict the MF ratings. There are cols like, 10 yr, 7 yr, 5 yr etc returns. I also have commencement date of MFs, the question is there are MFs with commencements dates only 3 yrs back, so would it be prudent to impute the returns for them for 10 yrs and 7 yrs period?

In my opinion that would be wrong, but then one of the requirements for ML models is to have no missing values in the data, and adding a 0 would be wrong too.

Need suggestions.

## Answer by ahron (score 1)

https://quant.stackexchange.com/a/53281

To put it very bluntly: The requirements of ML models are irrelevant to the principles of financial analysis.

Imputing the returns is fundamentally extremely unsound.

The best you can do is trim the dataset down to the smallest timeframe (in your case, 3 years). Or you can consider only those funds for which you do have sufficient 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.