Crude, Gold, Treasury Yields and VIX – Which one is most Predictive of Dow Jones Industrial Average?

-- Intended for New College Graduates --

(Click on the image to enlarge)

Julie is interviewing for an Equity Analyst position with a Wall Street Brokerage firm.

Question # 1
Interviewer: Julie, we used 13-months (i.e., 07/01/2018 thru 07/31/2019) worth of daily closing prices to compile this correlation matrix and the regression graph. Now, by looking at them, can you tell me what our objective here is?

Julie: You are trying to see if Gold, 10 and 30-year Treasury Yields, Crude and VIX collectively can predict Dow Jones Industrial Average (DJIA).

Question # 2
Interviewer: Why did we use ETFs like GLD and XOP instead of the actual futures data?

Julie: Futures contracts have different expiration dates so combining such data from different contract periods would be discontinuous. ETFs, instead, would be much better proxies.

Question # 3
Interviewer: In this example, is VIX the most un-correlated with DJIA? Qualify your answer with the underlying theory.

Julie: No. It's the most correlated of the five independent variables. Correlation can be positive or negative, hence the correlation coefficient varies between +1 and -1. VIX is negatively correlated with DJIA here.

Question # 4
Interviewer: In that case, which one is the least correlated independent variable here?

Julie: It's the crude ETF, that is the XOP variable in the equation.

Question # 5
Interviewer: Based on this correlation matrix, would you use all of the five independent variables in the regression equation? Qualify your answer with the underlying theory.    

Julie: No. I would remove GLD and 30-year Treasury Yield right off the top because they are failing the test of multi-collinearity. GLD is highly correlated with three others, while the 30-year Yield is moving in lockstep with the 10-year Yield.

Question # 6
Interviewer: Why did you choose 10-year Yield over 30-year Yield? Aren't they interchangeable here?

Julie: 10-year has better predictive relationship with the DJIA and lesser correlation with the VIX, while 30-year has only one positive, that is lesser correlation with XOP. Out of three, two positives here are better than one positive. Therefore, they are not necessarily interchangeable here.

Question # 7
Interviewer: The regression line shows a r-squared of 0.7633. What r-squared would the actual regression output show?

Julie: The same 0.7633. The regression value here represents all five independent variables against the same DJIA dependent variable so the r-squared would be identical. You are basically graphing the outcome of the actual regression.

Question # 8
Interviewer: If you are asked to fine-tune the model with an improved r-squared, what would you do? Qualify your answer with the underlying theory.

Julie: I would remove some outliers systematically from both ends of the curve. Unlike weekly closing prices, daily closing prices are inherently very volatile, so removing some outliers would be reasonable.

Question # 9
Interviewer: If you are forced to run a simple regression, rather than a multiple regression comprising these five variables, which one would you choose? And, what type of regression coefficient would you expect to see?

Julie: VIX, because it has the best predictive relationship with the DJIA. The regression coefficient would be negative as well, in line with the correlation coefficient.

Disclaimer - The author is not advocating the ETFs/indices listed here. Consult your Registered Rep, RIA or Financial Planner for an appropriate asset allocation model and the suitability of stocks, indices and other holdings for your portfolio.

Good Luck!

Sid Som, MBA, MIM
President, Homequant, Inc.
homequant@gmail.com

Coming soon: Sid's New Book: Modern Interviewing Techniques and Skills - Live Simulations with Actual Market Data

Does VIX Really Move in Tandem with S&P 500?

-- Intended for New Graduates --

VIX is the implied volatility index derived off the S&P 500, so it has become one of most widely watched and followed market metrics in the financial world, since its very inception. 

Many professional traders still define their market entry and exit based largely on the movement of the VIX and their  primary stock trading mantra continues to be:

"When VIX is high, it's time to buy.

When VIX is low, it's time to go."

So, let's use some recent market data (July 1, 2018 thru June 30, 2019) to examine if the S&P 500 index and VIX truly move in tandem and, if so, to what extent (meaning the extent of their statistical relationship).

As a student, you might have used the weekly closing data to establish such relationships, but now that you are ready to enter the corporate world, start making your case more emphatically with both daily and weekly closing data, where the daily serves as the "Champ" while the weekly "Challenges."

Though the correlation coefficient is the primary metric to derive and demonstrate such a relationship, a scatter plot with the trendline and the R-squared is as important considering it offers a more compelling visual case showing the line of best fit relative to the datapoints. 

In this instance, the correlation matrix of daily closings shows a high inverse correlation (-0.828) between them, meaning they move in tandem (not necessarily in lockstep) but in opposite directions. The daily scatter plot also confirms the same inverse relationship, with a fairly high R-squared (0.716). 

Therefore, the professional traders tend to use the VIX Options to hedge the S&P 500 Index (loosely, a long-short strategy).

As you expect, the correlation matrix of weekly closings would show a similar but smoother inverse relationship, and it certainly does (-0.842).

Likewise, the weekly scatter plot shows the same inverse relationship, as well as a negative trendline, predictably with a slightly tighter fit and a higher R-squared (rising to 0.747). 


PART-2


Now, let's simulate a job interview and frame a few meaningful questions out of the above presentation (remember, in a job interview they are not going to ask you some straight-forward questions like the definition of VIX or the S&P 500 Index, etc.)...

1. Interviewer: Megan, look at this Daily Scatter Plot and tell me if X and Y have been correctly graphed.

Megan: Yes, because VIX is a derivative of the S&P 500 index, and not the other way around. Using the basic math construct of Y is a function of X, VIX has been corrected depicted on the Y-axis.

2. Interviewer: Other than the two slightly different R-squared values, do you see any other difference between the Daily and Weekly Scatter Plots?

Megan: Yes, two basic differences: (a) obviously, the Daily plot has roughly five times more datapoints than the Weekly one, and (b) as expected, the Weekly Plot is smoother.

3. Interviewer: Do you see any technical inconsistency between the two scatters?

Megan: Yes, one. The X-scales are slightly different. They should have been held constant. 

4. Interviewer: Why do you think the Daily Correlation Coefficient is different from the Daily R-squared?

Megan: They are not apples-to-apples. The underlying maths are different. The Correlation Coefficient shows the overall statistical relationship between two variables, while R-squared shows to what extent (as a %) the independent variable explains the variations in the dependent variable.

5. Interviewer: As a follow-up to the prior question, why is the Daily Correlation Coefficient negative while the R-squared is positive?

Megan: Adding to my prior answer, the Correlation Coefficient can vary between +1 and -1, while the R-squared varies between 0 and 1 (cannot be negative), hence the difference. 

6. Interviewer: Why do you think we didn't show you any Regression output(s)?

Megan: Because it's a simple regression construct here, meaning one independent variable to the dependent variable. Had it been a multi-variate event, you would have produced a multiple regression output with the respective parameter estimates and the associated statistics.

7. Interviewer: You just indicated that had it been a multiple regression, we would have produced the respective parameter estimates with associated statistics. What sort of associated statistics would you have expected to see?

Megan: At a minimum, Standard Errors, T-stats and P-values.
   
8. Interviewer: Take a look at the two Scatter Plots and try to explain why a non-linear Trendline has been forced in.

Megan: Because of the slight tilt-up in the data at the outer end; I mean the most recent data seems to be bucking the trend a bit.  

9. Interviewer: Any guess as to the type of the Trendline?

Megan: Looks like, it's a 2nd or 3rd degree Polynomial.

Interviewer: Megan, we've a few more interviews this week so expect to hear back from us sometime next week. By the way, did you learn all this at school?

"No. My mom taught me."

Good Luck!

Disclaimer - The author is not advocating VIX or indices listed here. Consult your Registered Rep, RIA or Financial Planner for an appropriate asset allocation model and the suitability of stocks, indices and other holdings for your portfolio.


Sid Som, MBA, MIM
President, Homequant, Inc.
homequant@gmail.com