xcorr raw time series vs normalized time series

I have two time series which I have interpolated to be on the same dt. I am trying to compute the time lag between the time series. when I use xcorr on the raw time series, I get a lag of 0, whereas when i min/max normalize the time series first, then use xcorr i get a lag of -4. Why are the lags different when all I have done is normalize the data? Which one is correct?

DS(:,2) = -DS(:,2)

c1 = corr(DS(:,2),SS(:,2))

DSnorm = (DS(:,2)-mean(DS(:,2)))/std(DS(:,2))
SSnorm = (SS(:,2)-mean(SS(:,2)))/std(SS(:,2))

c2 = corr(DSnorm,SSnorm)

dt = 4;
[C,L] = xcorr(DS(:,2), SS(:,2),'coeff');
[~, idx] = max(abs(C));
L = L*dt;
Lag = L(idx);
plot(L,C)
hold on

[C,L] = xcorr(DSnorm, SSnorm,'coeff');
[~, idx] = max(abs(C));
L = L*dt;
Lag2 = L(idx);
plot(L,C)

Answer 1

[dpb]

DS=readmatrix('DS.csv');
SS=readmatrix('SS.csv');
DS(:,2) = -DS(:,2);

%c1 = corr(DS(:,2),SS(:,2))

DSnorm = (DS(:,2)-mean(DS(:,2)))/std(DS(:,2));
all(isnan(DSnorm))

ans = logical
   1

That's a problem -- all your normailzed D are NaN...

std(DS(:,2))

ans = NaN

You can try

DSnorm = (DS(:,2)-mean(DS(:,2)))/std(DS(:,2),'omitnan');
SSnorm = (SS(:,2)-mean(SS(:,2)))/std(SS(:,2));
all(isnan(SSnorm))

ans = logical
   0

%c2 = corr(DSnorm,SSnorm)

dt = 4;
[C,L] = xcorr(DS(:,2), SS(:,2),'coeff');
[~, idx] = max(abs(C));
L = L*dt;
Lag = L(idx);
%plot(L,C)
%hold on

[C,L] = xcorr(DSnorm, SSnorm,'coeff');
[~, idx] = max(abs(C));
L = L*dt;
Lag2 = L(idx);
%plot(L,C)
disp([Lag Lag2])

       -1056       -1056