[R-SIG-Finance] [R-sig-finance] Garch problem

[Patrick Burns]

The fit is essentially saying that the half-life
of a shock is infinite.  This generally occurs
when the in-sample volatility has a general
trend.  One solution is more data.  There are
doubtless other paths as well.

RON70 wrote:
> I have following dataset as monthly percentage return for a stock :
>
> 0.173741362
> -0.062237174
>
>   

[ ... ]
> -0.001652893
> -0.092301325
>
> Now I fit a GARCH (1,1) model on that :
>
>   
>> garch(Delt(dat)[-1], c(1,1))
>>     
>
>  ***** ESTIMATION WITH ANALYTICAL GRADIENT ***** 
>
>
>      I     INITIAL X(I)        D(I)
>
>      1     4.331103e-03     1.000e+00
>      2     5.000000e-02     1.000e+00
>      3     5.000000e-02     1.000e+00
>
>     IT   NF      F         RELDF    PRELDF    RELDX   STPPAR   D*STEP  
> NPRELDF
>      0    1 -4.507e+02
>      1    6 -4.508e+02  2.00e-04  3.20e-04  1.5e-03  6.3e+06  1.5e-04 
> 1.01e+03
>      2    7 -4.508e+02  1.57e-05  1.69e-05  1.4e-03  2.0e+00  1.5e-04 
> 3.19e-01
>      3   13 -4.521e+02  2.85e-03  4.72e-03  5.6e-01  2.0e+00  1.3e-01 
> 3.16e-01
>      4   16 -4.602e+02  1.76e-02  4.41e-03  8.1e-01  6.7e-01  5.1e-01 
> 1.99e-02
>      5   23 -4.607e+02  1.13e-03  2.77e-03  1.6e-04  7.4e+00  1.8e-04 
> 8.48e+00
>      6   24 -4.607e+02  4.81e-05  4.37e-05  1.6e-04  2.0e+00  1.8e-04 
> 1.77e+01
>      7   30 -4.638e+02  6.60e-03  8.81e-03  9.8e-02  2.0e+00  1.2e-01 
> 1.84e+01
>      8   31 -4.645e+02  1.52e-03  7.73e-03  8.2e-02  1.3e+00  1.2e-01 
> 1.39e-02
>      9   33 -4.688e+02  9.18e-03  6.28e-03  6.8e-02  0.0e+00  1.2e-01 
> 6.94e-03
>     10   35 -4.693e+02  9.32e-04  9.33e-04  8.9e-03  1.9e+00  1.8e-02 
> 2.86e-02
>     11   37 -4.699e+02  1.34e-03  1.59e-03  1.6e-02  1.8e+00  3.5e-02 
> 5.99e-02
>     12   38 -4.704e+02  1.05e-03  1.43e-03  1.6e-02  1.6e+00  3.5e-02 
> 9.10e-03
>     13   40 -4.705e+02  1.84e-04  2.85e-04  5.3e-03  1.2e+00  1.3e-02 
> 7.52e-04
>     14   42 -4.705e+02  3.71e-05  5.18e-05  2.4e-03  8.1e-01  5.0e-03 
> 7.09e-05
>     15   44 -4.705e+02  8.51e-07  3.04e-06  4.9e-04  8.2e-01  9.5e-04 
> 5.29e-06
>     16   57 -4.705e+02 -7.73e-15  1.09e-15  5.0e-15  4.4e+06  9.1e-15 
> 2.87e-07
>
>  ***** FALSE CONVERGENCE *****
>
>  FUNCTION    -4.704848e+02   RELDX        4.961e-15
>  FUNC. EVALS      57         GRAD. EVALS      16
>  PRELDF       1.088e-15      NPRELDF      2.867e-07
>
>      I      FINAL X(I)        D(I)          G(I)
>
>      1    2.824235e-05     1.000e+00     5.619e+01
>      2    8.649332e-02     1.000e+00    -5.899e-01
>      3    9.175397e-01     1.000e+00    -6.866e-01
>
>
> Call:
> garch(x = Delt(dat)[-1], order = c(1, 1))
>
> Coefficient(s):
>        a0         a1         b1  
> 2.824e-05  8.649e-02  9.175e-01  
>
> Warning message:
> In sqrt(pred$e) : NaNs produced
>
> What we see that sum of alpha and beta coef is more than 1. Therefore
> probably I choose a wrong model on my dataset. Can anyone please guide me
> how to modify that model?
>
> Regards,
>