Effacer les filtres
Effacer les filtres

How to calculate Sums in a Panel Data framework using "For Loops"

1 vue (au cours des 30 derniers jours)
John
John le 19 Déc 2016
Commenté : John le 21 Déc 2016
The following table consists of Price Returns returns at different Dates:
Idea: For any given Date and Stock Ticker, if returns data of previous 12 time steps is available (cell is not NaN), sum those returns and stock them in a new table. Code:
% Take row by row
% Considering columns 2 to 279 look at each date point if returns data for previous 12 time steps (weeks) is available
for i=length(Tnew.Date)
for j=TNew.x1COVGY:end
if (TNew.x1COVGY:end = ~(NaN))
% Sum those 12 entries and list the result in a new table. For example, A2AIM's value for 30-Oct will be the sum of values between 30-Oct and 14-Aug.
end
end
Thanks for your help!!
  2 commentaires
José-Luis
José-Luis le 19 Déc 2016
This looks very much like homework. What have you tried so far?
John
John le 19 Déc 2016
Hi José-Luis. For some it looks like homework, for others it is part of a larger research work, split into small pieces in order to understand the logic behind the codes. I tried the following example:
X = sum(Tnew.A2AIM)
but this summed up over the entire time period, ignoring the 12-time-steps frame I want to introduce. Also, this takes into account only 1 stock, whereas I cam trying to loop through all stocks. For creating a new table I know I can use the table() function.
Y = table(TNew.Date, ...)

Connectez-vous pour commenter.

Réponse acceptée

Chris Turnes
Chris Turnes le 20 Déc 2016
Modifié(e) : Chris Turnes le 20 Déc 2016
To get the moving sum of the previous 12 elements, and marking all windows NaN where you don't have 11 previous points, you can use the two-element window input [11 0] and the Endpoints name-value pair to movsum, like this:
>> movsum(1:15, [11 0], 'Endpoints', 'fill')
ans =
NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN 78 90 102 114
This is a trailing 12-point sum.
As far as dealing with all of the variables, you can always just extract them and reassign them from the table. movsum can deal with matrix inputs, and by default will go down the columns, so you could do something like this:
>> T = table(randn(20,1), randn(20,1), randn(20,1), randn(20,1));
>> T(1:3, :)
ans =
Var1 Var2 Var3 Var4
_______ _______ _______ _______
-1.2276 0.80352 1.0438 0.90102
1.1265 0.4427 -1.0934 0.33499
0.56487 -1.2102 1.3748 0.36737
>> T2 = T;
>> T2{:, 1:4} = movsum(T{:,1:4}, [11 0], 'Endpoints', 'fill')
T2 =
Var1 Var2 Var3 Var4
________ ______ ________ ________
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
NaN NaN NaN NaN
-1.3491 4.066 5.5826 -0.34325
0.35979 3.5101 4.6219 -1.7896
-0.62395 4.7247 5.7 -1.7089
-0.90211 6.4993 1.9725 -1.3553
-3.1805 8.3403 2.5239 -1.6156
-2.8086 7.9674 1.6148 -0.55556
0.84903 7.2745 2.9359 0.35856
2.3088 6.2673 0.084432 0.36349
3.2424 6.6688 -0.97282 0.57143
  1 commentaire
John
John le 21 Déc 2016
Very much appreciated! Thanks for your answer, Chris! This one works perfectly for my context.

Connectez-vous pour commenter.

Plus de réponses (0)

Catégories

En savoir plus sur Get Started with MATLAB dans Help Center et File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by