Fitlm results not obtainable

When I try to run the equation with two variables, one 83x370 (Rdec) and the other 83x8 (F) I get this error:

>> mdl=fitlm(F,Rdec)
Error using classreg.regr.FitObject/selectVariables (line 351)
Response variable must be a numeric vector.

Error in classreg.regr.TermsRegression/selectVariables (line 259)
            model = [e-mail](model);

Error in classreg.regr.FitObject/doFit (line 91)
            model = selectVariables(model);

Error in LinearModel.fit (line 1043)
            model = doFit(model);

Error in fitlm (line 121)
model = LinearModel.fit(X,varargin{:});

I have NaN values in my 83x370 but cant remove them whilst maitining the shape of the matrix. I have loaded the F and Rdec as numeric matrixes. Any help would be appreciated.

[John D'Errico]

You already asked the same question.

[Jeff Miller]

Before we get to the nan problem, please explain more about your variables. You said "When I try to run the equation with two variables, one 83x370 (Rdec) and the other 83x8 (F) ...". But fitlm sees Rdec as 370 variables (83 values each) and it sees F as 8 variables.

[John Good]

My F is my factor data of managers of 83 funds. My Rdec is the fund returns over 25 years for those 83 funds

[Jeff Miller]

What are the 370 variables in Rdec? Are these fund returns at various time points within the 25 years?

Answer 1

[Walter Roberson]

https://www.mathworks.com/help/stats/fitlm.html#bt0ck7o_sep_shared-y

The response variable must be a vector . Not a matrix.

If your intention is that you want to fit the 370 different responses for the same data, then you need to do so one response (column) at a time.

[John Good]

So if I split the returns into column vectors and then use an equation like this to go through them all would this work?

for i=1:funds; 
mdl=fitlm(F,Rdec(:,i)); 
alpha(i,:)=mdl.Coefficients.Estimate(1) 
talpha(i,:)=mdl.Coefficients.tStat(1) 
end;

[John Good]

I'm trying to determine if the 8 fund manager factors I have identified in F impact on the returns over the 25 years (Rdec). Maybe I am going about this the wrong way.

[Jeff Miller]

I think your 'for' loop would basically work, assuming 'i' would go from 1 to 370 (though I'm not sure what these 370 represent).

However, I would think that you would want to save coefficients for all 8 factors, not just the first one.

Given what you want to do, it seems to me that your best bet is to form some kind of composite measure of fund performance combining across all 370 variables, and then just look at a single regression predicting that composite from your 8 fund manager factors.

[Walter Roberson]

The loop looks plausible.