Clarification_editgrids_SRF_OLS_Multiple regression

[sdoshi004]

Hi David

Hope Alls well

In the captioned editgrids, though I have understood it conceptually, I could not understands few calculations like

SRF

1. calculation of b2
2. Variance of disturbance error
3. var (b1) and Se(b1) etc.

Similary in OLS- the calculation of reporting results and in Multiple regression - calculation of regression output.

Just want to know whether we are expected to perform such calculation or we have to only interpret these calculations.

If it is the former, than we would highly appreciate if you can do a half n hour session on the above spreadsheets to explain the math involved.

Would appreciate your response

Thanks

Sumit

 

[David Harper CFA FRM]

Hi Sumit,

We maybe crossed paths because last night I uploaded a 40-min review of the regression spreadsheets
(see http://www.bionicturtle.com/premium/screencast/2.b._spreadsheet_review_regression_volatility/)

It's great you have taken a close look; I think you understand exactly what you need to know if these three are difficult, you have identified the more difficult calculations and generally you will not (cannot) be expected to perform these calcs. Rather, in the review above, I highlighted the relevant calcs. In particular

1. b2 slope: it will not be tested as a raw calculation. But it is instructive to understand it. But please note, conceptually, this is a classic re-occurrence of "beta" as covariance()/variance(). If you think about beta CAPM, it is covariance(security, market)/variance(market); here is generically covariance(x,y)/variance(x).

2. variance of disturbance: it also will not be calculated. But, we do want to know that standard error (disturbance) = SQRT[RSS/d.f.]. At the same time, you can probably figure out this calc if you think about the regular variance (e.g., average of squared returns): in this case, it is "almost average" of squared residuals.

3. var(b1): no, you cannot be expected. In regard to se(b1), just know it is SQRT[var(b1)] and the more important application is using it to test the significance of the coefficient. So, the important formula is: (bn - Bn)/se(bn) follows student's t. Conceptually, we want to recognize that, just like the inference in Ch 5, the se(b1) is really just a standard deviation of b1, and we can use a student's t to test the significance of the estimate.

David