Handlining Missing qualtery financia data

sunnyal

Hi,

 

We are tying a use case to analyze and predict certain private companies (startups/well established) quarterly earnings. The data feed we have for some of the private companies do not have any financial information, specifically quarterly earnings. All that at the most we heave is their valuation. Is there any specific leaner that we can apply to determine the MISSING quarterly earnings and design approach for such need

 

The second part of the issue is some of our current private clients do send their quarterly earnings, but some have missing data, like they might have Q1, Q3, Q4 (missing Q2), what would be the best approach to handle missing data. I saw a an operator to handle missing data in RM, but is there a better way than simple average, min, max??

 

Thx

lionelderkrikor

Hi @sunnyal

 

I see a priori 2 methods : 

 

1.  Impute missing values operator : 

This operator estimates values for the missing values of selected attributes by applying a model learned for missing values.

 

2.Predict the missing values : 

Using the features which do not have missing values, you can predict the nulls with the help of a machine learning algorithm. This method may result in better accuracy, unless a missing value is expected to have a very high variance.

You can experiment with different algorithms and check which gives the best accuracy...it's kind of a machine learning problem inside the global machine learning problem

 

I hope it helps,

 

Regards,

 

Lionel

 

 

sunnyal

Thank you the challenge is that we only have valuation data of companies and predicting 4 quarterly earning out of that seem tricky. Perhaps how do we tell the model under supervised learning models that predict the quarterly earnings by learning (infering values) from similar companies which has similar valuation.

 

Is there a sample process I can infer please??

Thomas_Ott

@sunnyal I don't have a process created for something like this but it would be handy. My initial thought on how to do design this is get as much data as you can for demographics on companies you do know (i.e. size, valuation, earnings, employees, etc) and then build time series models, from there you can apply the companies you are interested in and find earnings?