Lately we have been experimenting with new financial visualization techniques. Yesterday we explored how violin plots could be used as a replacement for bar and candlestick charts. Today we are taking an introductory look at Twitter’s Breakout Detection package for the R statistical programming language. The package uses a statistical technique called E-Divisive with Medians to detect shifts in the mean of a time series. Twitter uses this technology to monitor it’s own network to detect changes in user activity.
We wanted to see how well the package could detect changes in the S&P 500 Volatility Index (VIX). The VIX tends to spike when stock markets decline, while the VIX compresses when stocks are rising or stable. Detecting shifts in the mean of the VIX is thus of great importance to a wide range of traders and investors.
We ran the daily closing value of the spot VIX through the breakout function. While more study is certainly warranted, a cursory glance shows that the function was able to identify a number of interesting moments in the VIX time series near volatility spikes. We are currently in a VIX breakout:
Twitter has also released an Anomaly Detection R package which incorporates mean-shifts into the detection algorithm. Future posts will investigate the use of both these packages in further detail.