Split data into quintiles

Good afternoon

I recently wrote a code (bottom of this question) for splitting a time series of return data (a different column for different stocks) into 5 portfolios depending on their betaCSD values. This worked perfectly. (To run the data, I copy the first line of the code into the command window).

I now want to repeat this but only for dates where the cross-sectional dispersion (CSD) is negative.

I split the CSD with its dates into two seperate matrices (where CSD is negative/positive), and then adapted all of the inputs to, also, only include values with respect to a negative CSD.

The inputs are: negcsdCSDdiff, betasdispnegcsdforcsddiff, NQ, thedatanegcsdforcsddiff, weightnegcsdCSDdiff, FFnegcsdCSDdiff

However, now, when I run the code, the outputs only appear to work for one row (the 5th) (e.g. outputRetnegcsdforcsddiff, attached).

Please could anyone see where I am going wrong?

Betasdispnegcsdforcsddiff and thedatanegcsdforcsddiff were over 3MB so I can’t attach them, so I have attached samples of them instead.

Thank you for your help

Emily

%% Commputes the returns of sorted portfolios

% Portfolios are rebalanced MONTHLY

%

% INPUT

% 'data' is a Nx(M+2) matrix of spot returns, structured as [datenum retMkt ret(1) ret(2) ... ret(M)]

% 'conData' is a Nx(M+2) matrix of data on which returns will be conditioned, structured as [datenum conMkt con(1) con(2) ... con(M)]

% 'weightData' is a Nx(M+2) matrix of values for constructing weights, structured as [datenum weightTOTAL weight(1) weight(2) ... weight(M)]

% 'NQ' is the number of quartiles at which data will be assigned

% 'CSAD' is a Nx2 matrix, structured as [datenum CSAD]

% 'FF' is a Nx3 matrix, structured as [datenum FF(1) FF(2)]

%

% OUTPUT

% 'outputRet' is a Nx(NQ+4) matrix of portfolio returns (mktCap-weighted), structured as [month year ret(1) ret(2) ... ret(NQ) ret(>0) ret(<0)]

% 'outputBetaDisp' is a Nx(NQ+4) matrix of portfolio CSAD betas (PRE-formation), structured as [month year beta(1) beta(2) ... beta(NQ) beta(>0) beta(<0)]

% 'outputBetaPost' is a Nx(NQ+4) matrix of portfolio CSAD betas (POST-formation), structured as [month year beta(1) beta(2) ... beta(NQ) beta(>0) beta(<0)]

% 'outputRetEW' is a Nx(NQ+4) matrix of portfolio returns(equally-weighted), structured as [month year ret(1) ret(2) ... ret(NQ) ret(>0) ret(<0)]

% 'alphasCAPM' is a Nx(NQ+4) matrix of portfolio alphas (POST-formation) based on the CAPM, structured as [month year alpha(1) alpha(2) ... alpha(NQ) alpha(>0) alpha(<0)]

% 'alphasFF' is a Nx(NQ+4) matrix of portfolio alphas (POST-formation) based on the FF model, structured as [month year alpha(1) alpha(2) ... alpha(NQ) alpha(>0) alpha(<0)]

% 'mktCapT' is a Nx(NQ+4) matrix of portfolio total mkt Cap, structured as [month year mktCapT(1) mktCapT(2) ... mktCapT(NQ) mktCapT(>0) mktCapT(<0)]

function [outputRetnegcsdforcsddiff, outputBetaPrenegcsdforcsddiff, outputBetaPostnegcsdforcsddiff, outputRetEW, alphasCAPM, alphasFF, mktCapT, FCSAD] = CSD_portfolioReturnsMONTHLY1forrobustness(thedatanegcsdCSDdiff, betasdispnegcsdforcsddiff, weightnegcsdCSDdiff, NQ, negcsdCSDdiff, FFnegcsdCSDdiff)

thedatanegcsdCSDdiff = sortrows(thedatanegcsdCSDdiff,1); betasdispnegcsdforcsddiff = sortrows(betasdispnegcsdforcsddiff,1); weightnegcsdCSDdiff = sortrows(weightnegcsdCSDdiff,1);

years = unique(year(thedatanegcsdCSDdiff(:,1))); template = []; for i = 1:size(years,1); a = [1:1:12]'; a(:,2) = years(i,1); template = [template; a]; end;

outputRetnegcsdforcsddiff = template; outputRetnegcsdforcsddiff(:,3:NQ+4) = NaN; % initialize output matrix for portfolio Returns

outputBetaPrenegcsdforcsddiff = template; outputBetaPrenegcsdforcsddiff(:,3:NQ+4) = NaN; % initialize output matrix for PRE-formation portfolio CSAD_betas

outputBetaPostnegcsdforcsddiff = template; outputBetaPostnegcsdforcsddiff(:,3:NQ+4) = NaN; % initialize output matrix for POST-formation portfolio CSAD_betas

outputRetEW = template; outputRetEW(:,3:NQ+4) = NaN; % initialize output matrix for portfolio Returns (equally-weighted)

alphasCAPM = template; alphasCAPM(:,3:NQ+4) = NaN; % initialize output matrix for CAPM alphas

alphasFF = template; alphasFF(:,3:NQ+4) = NaN; % initialize output matrix for FF alphas

mktCapT = template; mktCapT(:,3:NQ+4) = NaN; % initialize output matrix for Total Mkt Cap

betasFCSAD = template; betasFCSAD(:,3:NQ+2) = NaN;

for i = 2:size(outputRetnegcsdforcsddiff,1);

a = thedatanegcsdCSDdiff(year(thedatanegcsdCSDdiff(:,1)) == outputRetnegcsdforcsddiff(i,2),:); % find 'spotRet' for a given year

a = a(month(a(:,1)) == outputRetnegcsdforcsddiff(i,1),:); % find 'spotRet' for a given month within that year

aa = a(1,3:end)'; % keep all assets' return for that month in the first column of 'aa'

locX = find(thedatanegcsdCSDdiff(:,1) == a(1,1));

aa(:,2) = betasdispnegcsdforcsddiff(locX,3:end); % keep assets' conditioning data for that month (2nd column)

aa(:,3) = weightnegcsdCSDdiff(locX,3:end); % keep assets' weighting data for that month (3rd column)

aa(:,4) = [1:1:size(aa,1)];

aa = aa(~isnan(aa(:,1)),:); aa = aa(~isnan(aa(:,2)),:); aa = aa(~isnan(aa(:,3)),:); % drop NaN observations

aa = sortrows(aa,2); % sort all data according to conditioning data (i.e. according to their CSAD beats)

cc = negcsdCSDdiff(year(negcsdCSDdiff(:,1)) == outputRetnegcsdforcsddiff(i,2),:); % find CSAD observations for that month

cc = cc(month(cc(:,1)) == outputRetnegcsdforcsddiff(i,1),:);

ff = FFnegcsdCSDdiff(year(FFnegcsdCSDdiff(:,1)) == outputRetnegcsdforcsddiff(i,2),:); % find pricing factors observations for that month

ff = ff(month(ff(:,1)) == outputRetnegcsdforcsddiff(i,1),:);

if ~isempty(aa)

dataX = thedatanegcsdCSDdiff(year(thedatanegcsdCSDdiff(:,1)) == outputRetnegcsdforcsddiff(i,2),:); % find 'data' obs for a given year

dataX = dataX(month(dataX(:,1)) == outputRetnegcsdforcsddiff(i,1),:); % find 'data' obs for a given month within that year

if size(dataX,1) >= 15 % check that at least 15 returns are available for a given month

Nrows = floor((size(aa,1)-2)./NQ);

for j = 1:NQ+2

if j < NQ

b = aa((j-1)*Nrows+1 : j*Nrows,:); % get b = [ret beta mktCap A/A] for portfolio j

elseif j == NQ

b = aa((NQ-1)*Nrows+1 : end,:); % get b = [ret beta mktCap A/A] for last portfolio

elseif j == NQ +1

b = aa(aa(:,2) > 0,:); % get b = [ret beta mktCap A/A] for portfolio with positive betaDisp

elseif j == NQ+2

b = aa(aa(:,2) <= 0,:); % get b = [ret beta mktCap A/A] for portfolio with negative betaDisp

end

w = b(:,3)./sum(b(:,3)); % compute (percentage) weights for each stock

outputBetaPrenegcsdforcsddiff(i,j+2) = sum(w.*b(:,2)); % compute PRE-formation beta as the weighted average of stocks' betas

mktCapT(i,j+2) = sum(w.*b(:,3)); % compute Total Mkt Cap as the weighted average of stocks' mkt caps's

bb = NaN(size(dataX,1),size(b,1)+2); bb(:,1) = dataX(:,1);

for k = 1:size(b,1)

bb(:,k+2) = dataX(:,b(k,4)+2); % identify returns' time-series for asset k within given month

end

bb(isnan(bb)) = 0;

bb(:,2) = bb(:,3:end)*w;

% bbret = bb(:,2); % avoid error in geomean 1/3

% bbret(bbret<=-1) = -0.9999; % avoid error in geomean 2/3

% bb(:,2) = bbret; % avoid error in geomean 3/3

outputRetnegcsdforcsddiff(i,j+2) = mean(1+bb(:,2))^size(bb,1)-1;

slopes = regress(bb(:,2), [ones(size(bb,1),1) a(:,2) cc(:,2)]);

slopesCAPM = regress(bb(:,2), [ones(size(bb,1),1) a(:,2)]);

slopesFF = regress(bb(:,2), [ones(size(bb,1),1) a(:,2) ff(:,2) ff(:,3)]);

slopesFCSAD = regress(bb(:,2), [ones(size(bb,1),1) cc(:,2)]);

outputBetaPostnegcsdforcsddiff(i,j+2) = slopes(2,1);

alphasCAPM(i,j+2) = slopesCAPM(1,1);

alphasFF(i,j+2) = slopesFF(1,1);

outputRetEW(i,j+2) = mean(bb(:,2));

end

b = aa((NQ-1)*Nrows+1 : end,:);

w = b(:,3)./sum(b(:,3));

end

end

end

FCSAD = template;

for i = 1:size(FCSAD,1)

FCSAD(i,3) = sum(betasFCSAD(i,3:7) .* outputRetnegcsdforcsddiff(i,3:7));

end