correlation error: 'x' must be numeric

Question

I have a XTS dataset that contains many stock closing prices called: dataset. I then wanted to find if their returns have any correlation via cor() , however I get an error message: Error in cor(RETS) : 'x' must be numeric.

Here is what i have done:

RETS <- CalculateReturns(dataset, method= c("log")) # Calculate returns Via PerformanceAnalytics
RETS<- na.locf(RETS) #Solves missing NAs by carrying forward last observation
RETS[is.na(RETS)] <- "0"  #I then fill the rest of the NAs by adding "0"

Here is a sample of RETS

    row.names   A.Close    AA.Close AADR.Close  AAIT.Close   AAL.Close
1   2013-01-01    0            0            0         0         0
2   2013-01-02  0.0035      0.0088      0.0044      -0.00842    0
3   2013-01-03  0.0195      0.0207     -0.002848    -0.00494    0
4   2013-01-06 -0.0072     -0.0174      0.0078      -0.00070    0
5   2013-01-07 -0.0080      0          -0.01106     -0.03353    0
6   2013-01-08  0.0266     -0.002200    0.006655     0.0160     0
7   2013-01-09  0.0073     -0.01218     0.007551     0.013620   0

Then I perform the correlation:

#Perform Correlation
cor(RETS) -> correl
Error in cor(RETS1) : 'x' must be numeric

#Tried using as.numeric
cor(as.numeric(RETS), as.numeric(RETS) -> correl

However the answer is "1". I also tried using the correlation function in psych but get the same error message.

Goblin

Could you show us the result of typeof(RETS) ? – Obduliaobdurate

@Obduliaobdurate Yes, typeof(RETS) = "character" – Goblin

What is your question? By using RETS[is.na(RETS)] <- "0" you turn all your data into characters and you can't calculate a correlation of characters. – Heaney

+1 to @Heaney comment. You should do RETS[is.na(RETS)] <- 0. Moreover correl should be matrix because your data.frame RETS has several columns – Obduliaobdurate

@Heaney Oh now i understand. That makes it work now, thanks guys! – Goblin

Answer 1

I'm adding @Roland's answer where to close out the question.

The problem is that using

RETS[is.na(RETS)] <- "0"

is turning all the data into characters since adding any character value to a numeric value automatically changes the data.types to a character. Thus when you go to take the correlation, there is no way to do that for character values. So if you simply do

RETS[is.na(RETS)] <- 0

instead, you should avoid the conversion problem.

Rather than setting your missing values to NA, you might also consider explicitly telling cor how to handle missing values For example

cor(RETS, use="pairwise.complete.obs")

will only calculate correlation between two variables for those pairs where both are not-NA. See the ?cor help page for all of the options.

Weig