Correlation VXX vs XIV, Year 2011

Chart 1 Correlation %XIV vs %VXX, Year 2011
showing good, negative, linear correlation (R squared = 0.994)

Correlation %XIV vs %VXX, Year 2011
Click charts to enlarge

Sum 20.69 -32.38
Days 252 252
Average 0.08 -0.13
StDevP 4.66 4.63

Chart 2 Correlation VXX vs XIV, Year 2011 
showing some interesting patterns and voids
and pretty good, negative, exponential correlation (R squared = 0.914)
but not as good as Chart 1

Correlation VXX vs XIV, Year 2011

Chart 3 showing VXX vs XIV, Year 2011 by month

Chart showing VXX and XIV by month, Year 2011

Chart 4 showing VXX vs XIV, Year 2011 by month

Chart VXX vs XIV by month, Year 2011

Chart 5 showing VXX vs XIV, Year 2011

Chart VXX vs XIV, Year 2011

VXX XIV ratio
Sum 8418 2976 964
Days 252 252 252
Average 33.40 11.81 3.83
StDevP 8.90 3.95 2.67

Chart 6 Google Finance Chart XIV vs VXX, Year 2011
showing percentage changes from Jan 3, 2011

Charts 4, 5 and 6 show that both VXX and XIV are not good long-term hedges against volatility, especially XIV unless it's shorted, which was down 46% last year:

Dec 30, 2011 35.53 6.51
Dec 31, 2010 37.61 11.95
% Decrease 6% 46%

Click here to read an article that also concludes its better to short VXX than go long XIV, as Charts 4,5 and 6 suggest too.

It may seem odd that VXX and XIV cross each other near zero % twice, then diverge with XIV down 46% and VXX down only 6% at year end. The primary reason for this divergence can be seen in Charts 4 and 5 — in the last 5 months of 2011, increased VIX volatility caused VXX price to increase significantly from its low, and XIV price to decrease significantly from its high. Contango contributed to decline prices.

The exponential regression equation

For exponential trend lines a transformation to a linear model takes place. The optimal curve fitting is related to the linear model and the results are interpreted accordingly.

The exponential regression follows the equation y=b*exp(a*x) or y=b*m^x, which is transformed to ln(y)=ln(b)+a*x or ln(y)=ln(b)+ln(m)*x respectively.

a = SLOPE(LN(Data_Y);Data_X)

The variables for the second variation are calculated as follows:

m = EXP(SLOPE(LN(Data_Y);Data_X))

b = EXP(INTERCEPT(LN(Data_Y);Data_X))

Calculate the coefficient of determination by

r² = RSQ(LN(Data_Y);Data_X)

Besides m, b and r² the array function LOGEST provides additional statistics for a regression analysis.



Click to enlarge

Related Tickers: VelocityShares Daily 2x VIX ST ETN (NYSEArca:TVIX),  VelocityShares Daily Inverse VIX ETN (NYSEArca:XIV), VelocityShares Daily Inverse VIX Medium-Term ETN (NYSEArca:ZIV), iPath S&P 500 VIX Mid-Term Futures ETN (NYSEArca:VXZ), VIX Mid-Term Futures ETN (NYSEArca:VXZ), iPath S&P 500 VIX Short-Term Futures ETN (NYSEArca:VXX)