When we forecast financial markets (such as stock returns), we may use a variable without transformation (i.e., level). However, a growing body of research uses first difference, log difference, regression residual, regression slope, and time trend instead of the level. That is, markets react to new or future information (innovations) instead of the expected information (such as the level). Please provide more explanations and discussions.
Muhammad Abubakr Naeem can you please help in solving the problem here?
Hi, stock prices, commodity prices and the alike are usually well described by unit-root processes. The same happens with the natural logarithm of these prices. In such case, a crude analysis involving variables in levels or log levels may be spurious. Please google "spurios regressions" for more details. Even the computation of simple correlations between trending variables or variables with unit roots may lead to the illusion of a strong linkage when in reality there is no relationship whatsoever.
When you transform the series to work with returns that do not display a trending behavior, the problem of a false discovery is minimized.
So I would say that in many cases researches avoid to work with series in levels to avoid the porblem of spurious regressions and correlations.
Hope it helps!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression