I am in the process of using a regression model to help predict the forecast of soft drinks. I have 52 sets of weekly data and my independent variables are feature space, temperature, price, competitor price and competitor feature space after having omitted some predictor variables due to multicollinearity or insignificant results.

I now have a model at a 95% confidence level where all p values are below 0.05, adjusted r squared is around 90. However my standard error is very high! In a collection of data which spans from around 100,000 - 450,000, the mean value being about 150,000, my standard error is 30,000. I am therefore reluctant to trust the accuracy of the model when making predictions. I have tried taking 2 years worth of data instead so doubling the sample size but it hasn't helped at all. Any ideas/advice?