Missing data reduce the representativeness of the sample and it distorts inferences about the population. Moreover, some researchers act as if the analyses of variables with missing data will still be correct on average, but this may be untrue. Although, there are several methods of dealing with missing data problem but the best method appears unknown. What is the best method to solve missing data problem when conducting empirical research in finance and economics (using either time series data or panel data)? Your contributions are welcome.
There is no best method as such. The usefulness of each method depends on the nature and extent of missing data. Use of a moving average could be sufficient for data that trends in a predictable manner. Use of an arithmetic mean could also relatively work for cross-sectional data. A more rigorous approach I would recommend is multiple imputation using SPSS software. The good aspect of this technique is that it imputes according to the nature of the missing data. For instance, if it is binary it uses a logistic approach. Where missing data is either semi-annual or quarterly data, cubic spline could be the appropriate technique. All these techniques are elaborated in the literature.
Depending on the application and on the way that performance is measured (statistical or economic performance, for example), different methods could result to be the best ones. I suggest to try using methods that are able to deal with missing values by constructing so called surrogate variables; among them, one method that is very popular and well performing are regression of classification trees (for example the CART approach). In different applications in eonomic and finance trees have already shown to be quite strong for prediction.
There is no "best" method, it depends on particular data/application. Sometimes I would use imputed values (e.g. in time series previous value), sometimes I would just delete some observations. For example: I have daily data for ten stock prices. Consider these cases:
1. On one particular day, there is missing value for one of the stocks. In this case I would simply use value from previous day.
2. On one particular day, there is missing value for most of the stocks. In this case, I drop that day from my data.
Moreover, the optimal way how to handle this problem depends not only on the data, but also on the type of the problem you want to study. Sometimes it is ok to use imputed values, sometimes it is very inappropriate.
missing data is a problem for econometric analysis especially in measuring heteroscedasticity but you can substitute a previous day for the missing data to fill up (when using daily/weekly data) but there is no best method. Many methodologies can give statistical inference but the result may not be reliable
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series