Rolling correlation in pandas returning NaNs

Question

I would like to get the best match of a small array "b" (length: few hundreds of elements) from a bigger array "a" (length: few millions of elements). I am trying to use pandas, rolling and corr for comparing "b" with a sliding window over "a". This is my code:

import pandas as pd
    
a = pd.read_csv(<file1>) 
b = pd.read_csv(<file2>)
    
normalized_a = (a - a.mean()) / a.std() 
normalized_b = (b - b.mean()) / b.std()

res = a.rolling(window=len(b)).corr(b)

Dataframe a is:

                 0
0        0.941042
1        0.656281
2        0.969081
3        0.881595
4        0.848359
...           ...
1814386 -1.323574
1814387 -1.351035
1814388 -1.359450
1814389 -1.296941
1814390 -1.266813

Dataframe b:

0   -2.256496
1   -2.949674
2   -1.614618
3   -1.784006
4   -0.976331
..        ...
287  0.378578
288  0.247859
289  0.375981
290  0.444575
291  0.450435

However, res contains all NaNs, but one element (in fact, output of res.count() is 1):

          0
0       NaN
1       NaN
2       NaN
3       NaN
4       NaN
...      ..
1814386 NaN
1814387 NaN
1814388 NaN
1814389 NaN
1814390 NaN

The only non-NaN element in res is located at row 291 (found with res.idxmax()):

280       NaN
281       NaN
282       NaN
283       NaN
284       NaN
285       NaN
286       NaN
287       NaN
288       NaN
289       NaN
290       NaN
291 -0.134144
292       NaN
293       NaN
294       NaN
295       NaN
296       NaN
297       NaN
298       NaN
299       NaN

Does anybody know why I get all these NaNs? I would have expected to get meaningful values after row 292. Is corr a pairwise operation?

Thanks!