Why do I get NaN values in linear regression?

I run the following regression:

mdl=fitlm(Reg_var(1:end,2:end),Reg_var(1:end,1),'RobustOpts','on')

My output:

 Linear regression model (robust fit):
    y ~ [Linear formula with 21 terms in 20 predictors]

 Estimated Coefficients:
                    Estimate          SE          tStat       pValue 
                   ___________    __________    _________    ________

    (Intercept)      0.0015684    0.00082514       1.9008    0.057371
    x1              -0.0025121     0.0011669      -2.1528    0.031375
    x2              -0.0011949     0.0011669      -1.0239      0.3059
    x3              0.00021162     0.0011669      0.18135      0.8561
    x4              -0.0015328     0.0011669      -1.3135     0.18905
    x5              -0.0012053     0.0011669      -1.0329      0.3017
    x6              -0.0020596     0.0011669      -1.7649    0.077622
    x7              -0.0032643     0.0011669      -2.7974    0.005168
    x8              0.00031066     0.0011669      0.26622     0.79008
    x9              -0.0021079     0.0011669      -1.8063    0.070913
    x10            -0.00046861     0.0011669     -0.40158     0.68801
    x11            -0.00037094     0.0011659     -0.31816     0.75038
    x12            -0.00017788     0.0011669     -0.15243     0.87885
    x13            -5.3013e-05     0.0011669    -0.045429     0.96377
    x14            -0.00072536     0.0011669      -0.6216     0.53423
    x15                      0             0          NaN         NaN
    x16            -0.00057762     0.0011669       -0.495     0.62062
    x17             -0.0003968     0.0011669     -0.34004     0.73384
    x18            -0.00093411     0.0011669     -0.80049     0.42346
    x19             -0.0017195     0.0011669      -1.4735     0.14065
    x20             -0.0015061    0.00096437      -1.5618     0.11839

 Number of observations: 6260, Error degrees of freedom: 6240
 Root Mean Squared Error: 0.014
 R-squared: 0.0043,  Adjusted R-Squared 0.00127
 F-statistic vs. constant model: 1.42, p-value = 0.107

Why do I get NaN values at dummy 15 and what does this mean? How can I solve this? What am I doing wrong?