When training models it is always wise to remove outliers, but it is also important to assess why they are outliers. Removing all the data that does not fit your desired trend and calling them outliers is not the way to go. But identifying common measurement problems or problems of other sources and explaining why they cause the appearance of outliers in your data makes you understand the system better and therefore build better models.

In a data limited scenario it is not advisable to ignore the outliers. But it is better come for rationale decision after analysis the probility of the catch.

Ragavan Nadarajah first you have to define how you want to assess the stock (what KPIs) then in general you can follow the following steps to conduct data cleaning:

1- Inspect all of your data, and make sure to detect any missing values, anomalies, and inconsistencies in the data.

2- Handle missing data.

3- Remove duplicates.

4- Data Type Conversion: Ensure that each variable is in the correct data type.

5- Consistent Measurement Units.

6- Outlier Detection.

7- Data Transformation: Depending on the requirements of your analysis, you may need to transform the data.

8- Data Normalization: Depending on the objectives of your analysis, consider normalizing the data.

These steps only clean your data not involve analysis.

For the outliers, first, investigate the points and if there is any pattern of the outliers' points, and you may consider using robust statistical methods that are less sensitive to outliers, such as median-based estimators.

Removing them without investigation may lead to bias and incorrect results.