high level statistics stuff

[skyhawk]

hello. 3 suggestions. I) use eviews to perform a principal component analysis - very simple to execute with the software. ii) use error correction (and cointegration to test validity of long run trend) iii) if you want to keep things super simple, just use correlation analysis. depending on your time frame, you could use either 3m or 12m rolling correlations to get a sense of which factor is generally strongest and crucially to get a sense of the dynamics. eg factor 1 might have a long run correl of 0.8, but has a tendency to oscillate seasonally and even go negative. hope that's useful.

 

[skyhawk]

forgot to add - regarding which factors might lead, you will need to shift the comparison time series by whatever lead/lag factor you are using (eg 3mths) and retest the correlations. you need to use some intuition initially but then you can easily see which correlation is strongest. bear in mind this is relatively static and has limitiations, but gives you a decent cursory insight. I recently did this exercise for test pmis, ips etc for CHina gdp so I think it is worth doing before conducting more rigorous econometric tests (if you are heading in that direction). helps to simplify initial regression models too, otherwise you run risk of just data mining

 

[dtrader98]

Suppose I have 5 variables that I hope will predict, or at least correlate with a sixth and seventh.
1) I want to find out how to weight them.
2) But also, I suspect that some of the variables might "weigh" more when other variables are high or low.
3) In addition some of the variables may have a predictive lead that is more or less (in time) than the others.

I have an idea how to go about doing it in a tedious fashion in excel, but does anyone know of any software or plugins that do all this in a more automated fashion?

As was previously mentioned (GAT) python, matlab, and R are all great alternatives to Excel for this type of analysis. If you can learn and utilize any of them, you'll seldom find reason to go back to Excel for modelling.

1) Many ways to do this. Pretty much the goal of statistical modelling.
Regression is one type of modelling and the classical method is to build a simple linear regression model.
2) One approach is to use variable categorical factors for 'high' and 'low' in your model. The model should weight other variable weights proportional to the factors. Look into mixed categorical and continuous predictor type models. You could also just leave them as continuous and let the model find the proportional weighting based on the data.
3) Time series is a good area to look at lead-lag relationships (Again, see GAT -- VAR, VECM).
But you could also just add lagged variables as input factors (time delay embedded matrix).

A suggestion would be to find a good introductory statistics or data science book centered around R, matlab, or python and work through the examples. Lots of free online courses as well.