Tag: VIX

Previously, we explored using Meta’s Prophet library to predict VIX. It turned out that simply extrapolating the last value of the index worked better (Prophet for VIX). In fact, locf (last one carried forward) works better than GARCH(1, 1) and most other approaches. Can the same be said about Conditional Gaussian Mixture Models?

We used the cgmm python library to forecast 20-day forward VIX and compared its root-mean-squared errors (rmse) to those of locf‘s.

locf is pretty hard to beat.

Especially so when the VIX index itself is volatile.

Code up on github.

Open sourced by Meta back in 2017, Prophet is a procedure for forecasting time series data. How does it compare to GARCH(1,1) and locf (last one carried forward) for forecasting VIX 20 days out?

We fit 500-days of rolling VIX data using Prophet and GARCH(1,1) and forecast forward 20-days. We then calculate the RMSE (Root Mean Squared Error) of the forecast vs. actual of both the models and locf. Plot RMSE of all three.

Ideally, you want the error to be low and the tail of errors to be as short as possible. GARCH(1,1) looks worse than Prophet. However, locf beat both?

When in doubt, take the average.

Code on github.

Also: VIX Seasonality

India VIX is a volatility index computed by NSE based on the order book of NIFTY Options. For this, the best bid-ask quotes of near and next-month NIFTY options contracts. India VIX indicates the investor’s perception of the market’s volatility in the near term i.e. it depicts the expected market volatility over the next 30 calendar days. Higher the India VIX values, higher the expected volatility and vice versa. (NSE)

Does the actual volatility come close what the VIX was implying 30 calendar days before? Not always and probably never.

What if it’s pricing something more immediate? Here’s the regression with a 10-day lag:

Regression with no lag:

The relationship between implied and historical is one of those things that are directionally true… sometimes.

Code and charts on github.

Is India VIX seasonal? Yes.

There is a huge amount of dispersion in the daily data when grouped by months. Taking averages of these may not make much sense.

However, when you decompose the series, you get some interesting monthly seasonality.

Zooming into the “season_year” chart:

If you transform the seasonality component and plot it by month, you’ll notice why everybody gets nervous in May.

Code and charts on github.