I really need some help here since I feel I have no one to ask this.

I am doing a research to find the relationship between the amount of Weibos (Chinese Twitter/Social Media) on a subject and the volatility on market prices, as keywords that the Weibo contain I chose 'gold' and 'stock'.

I got 2 samples, one around 30 days with a little over 4000 weibos and one around 38 days with a little over 6000 weibos, they are not random samples, in which I got them from the 100 most influential (by follower count) finance related accounts in weibo.

At one point my teacher asked me to check for stationarity and do a regression, however I had never done anything besides simple regressions.

So I start to use Eviews for this, do the unit root panel test, to which I find that both my samples, on both keywords, the vast majority of tests find a p lower than 1%, this means I can do a linear regression right?

I do ordinary least squares regression but the p I find for most samples is quite high, both stock keywords sample regress to the stock price index volatility (absolute change in price) with a p of 0.22 and 0.35 (sample 1 and 2), gold absolute market price change against gold keywords in weibo regresses at 0.32 and 0.01 in sample 1 and 2.

I try on the general linear model (GLM) functions, and I notice that if I choose newey west covariance method, family normal, link identity, my p value lowers considerably.

Is it ok for me to use the newey west covariance method in my case? I am a begginer and any tips or pointing in the right direction on the regression that can/should be used in my case would be very much appreciated!

Thank you very much, Daniel


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.