Prediction intervals for OLS's model.predict() without using wls_prediction_std

[James]

I feel I might be missing something obvious here, but is there an easy way to output a model's prediction with PIs for some new input without using wls_prediction_std? I am using a very simple OLS model, code to help my explanation is here:

model = sm.OLS.from_formula('mpg ~ horsepower', df)
model = model.fit()
x_pred = pd.DataFrame(data=[98],columns=['horsepower'])
print(model.predict(x_pred))

This yields a single predicted value, however I am interested in knowing the values at the 95% prediction interval. As mentioned I know there is a way to do this with wls_prediction_std as follows:

bias = pd.Series(np.ones(shape=(len(x_pred))))
x_pred = pd.concat([bias,x_pred],axis=1)
print(x_pred)
print(wls_prediction_std(model, exog=x_pred.values))

However this clearly doesn't work very seamlessly with the OLS functionality (e.g. compared to the methods of RegressionResults). Is there a method in RegressionResults that I've missed?

[josef...@gmail.com]

There is a not well advertised and maybe still experimental method that works the same as predict but returns a prediction results instance with standard errors and confidence and prediction intervala.

http://www.statsmodels.org/devel/generated/statsmodels.regression.linear_model.RegressionResults.get_prediction.html

Josef