0 and -inf in the residuals inferred from a ARIMA model

Hello, everyone.

when I was inferring residuals and variance from a estimated ARIMA model, I found some -inf and 0 in the residuals and variance and I don't know what's wrong. This error doesn't happen regular and it only happened in certain dataset. For example, when I infer the data Rtn in the attachment.

My codes are

options   = optimset('fmincon');
options   = optimset(options, 'Display'  , 'off', 'Diagnostics', 'off', ...
    'Algorithm', 'sqp', 'TolCon'     , 1e-7);
model= arima('AR', NaN,'ARLags',1, 'Distribution', 't', 'Variance', egarch(1,1));
fitmodel = estimate(model, Rtn, 'print', false, 'options', options);
[residuals, variances] = infer(fitmodel, Rtn);

% Calculate the Standardized Residuals
residuals = residuals ./ sqrt(variances);

When I look at the residuals and variances vector, I find some -inf and 0. I don't know why this happened. But, when I change another data, the infer function works well. Does anyone can help me to figure out this problem.

Thanks, very much.

Answer 1

[Roger Wohlwend]

For an EGARCH(1,1) model the ARCH and GARCH coefficients are expected to be positive. In your case they are not. That is why you have problems with the residuals and the variances. It seems that an EGARCH model is not appropriate for your data. I don't know the reason for that, I guess nobody can tell you. So the only way to fix this is to use a different model for the variance: GARCH(1,1) for example.

[Taylor Xie]

Thanks very much for helping me find the problem.