Thinking

Volatility Series 2

Pearson_Phil
November 16, 2015 Philip Pearson
Volume Profiles

In a continuation of our volatility series (last we looked at the trading halts, specifically around August 24th), this week we take an in-depth look into volume profiles. Given the high volatility period of August and September 2015, many questions were raised about how price volatility affects the markets. One of the most interesting is whether high volatility days have a dramatic effect on when trading takes place. Said differently, how do these days change the typical VWAP “smile” curve? Most traders assume that the volume profile differs on days of high volatility, and that is a hypothesis which we want to test.

Using data from 4/1/15 through 10/16/15, we categorize each day in the following fashion, comparing the S&P 500 opening price to that of the previous day’s closing price.

  • “Up” – Days in which the S&P 500 opened more than 0.5% higher than the previous day’s close.
  • “Down” – Days in which the S&P 500 opened more than 0.5% lower than the previous day’s close.
  • “Neutral” – Days in which the S&P 500 opened between 0.5% up and 0.5% down versus the previous day’s close.
  • August 24th – we took out August 24th because of the extreme volatility and used it as a side-by-side comparison with the other data sets.

S&P 500 VOLUME DISTRIBUTION

We test not only whether price volatility (a large difference in opening prices compared to the previous day) affected the volume profiles of the market, but also whether the direction of the move makes a difference. We used overall volume for the S&P 500 to build the volume profiles. During this period, there were 22 down days (21 excluding August 24th), 20 up days, and 96 neutral days.

The below data is broken down into 15-minute bins with the first bin being the opening print and the last bin being the close. While they seem similar, there are a few interesting takeaways. As seen on the table below, the most extreme divergence takes place with the opening and closing auctions. On days where the market is down, there was almost twice as much volume (2.43%) done in the opening auction vs. 1.23% for neutral days and 1.17% for Up days. Additionally the close was almost 2 percentage points larger proportionally for the days where the market opened down 0.5% or more. One possible reason: many of the down days see large intraday price swings, making the stability of the closing auction attractive.

ITG_Pearson_Figure1_Volatility_2015_HTML

August 24th saw a slightly different pattern from the other down days. While the opening auction was only slightly heavier than usual proportionally, there was a notable increase in trading over the first hour. This is partially due to all of the stock halts and resumptions that occurred.

Interestingly, there was very little difference between days where the market opened substantially up versus days where there was not much pre-market volatility. There is definitely more volume being traded when there is a lot of price volatility, high volume days in August saw almost double the total volume being traded, yet the shape of volume seems to stubbornly stick to the typical VWAP-curve.

ITG_Pearson_Figure2_Volatility_2015_HTML

INDIVIDUAL STOCK PROFILES

Additionally, we looked at when there was extreme movement in individual stocks. What happens when a stock is up or down > 5% from the previous day? This happens even in times of generally calm markets whether there’s an earnings release, news event, or research report comes out. We used a much higher threshold here because there are more extreme events on a single-stock level. Also, as the control, we only use stocks which open within +/- 1% of their previous close to represent a “neutral” day. Additionally, we narrowed the time period to 8/20 – 10/16, in an effort to focus on the period of recent volatility. One interesting note: over 90% of the instances of S&P 500 stocks opening more than 5% below their previous close (over a two month stretch) occurred on August 24th. Because of this, we excluded August 24th from the below data set. There were approximately 50 instances for each Up and Down 5%, compared with 13,693 instances where stocks opened within +/- 1% of their previous close.

ITG_Pearson_Figure3_Volatility_2015_HTML

The first 60 minutes of the day sees a much larger proportion of trading on stocks that have moved substantially overnight. This is nearly identical regardless of the direction of the move. In the first hour, stocks that opened up > 5% traded almost 30% of the day’s volume. This effect was even more dramatic for stocks opening down 5%, where that figure was 33% of the day’s volume. These are both about double the 16% seen in the neutral group. This morning-effect continues to persist until around 11:00. On the high volatility names, there is a much smaller percentage done at the end of the day and in the closing auction, which implies a reversion to average levels of trading volume once price discovery is established.

CONCLUSION

Overall, we expected to see much larger differences in market-wide volume profiles for high volatility days compared with the average trading day. Even the extreme volatility of August 24th did not create a very large deviation from the historical volume profile. We infer from this result that market-wide volatility has an obvious effect on the level of volume, but not the slope of the volume level. This contrasts plainly with the occasions of high single-stock volatility. When stock-specific news causes a stock to move significantly overnight, a clear pattern emerges, featuring heavy trading in the morning.

What does this mean for VWAP algorithms? In VWAP, you want to trade your order in exact proportion to the market. In practice, it is very difficult to tell if a higher than average level of observed volume will persist (meaning you should trade the historical schedule) or revert (meaning you should trade more now). One quick inference from the data in this report is that VWAP tracking may improve if one front-loads on high single-stock volatility days, while front-loading on market-wide volatility alone will result in poor performance.

All volatility results in increased trading levels. There becomes a clear divergence: will the volume level revert relatively or will it persist via higher levels through the entire trading day? Our intuition was that trading levels would be proportionally higher in the morning for both of the scenarios we looked at, but this was proven incorrect for the scenario of market-wide volatility. Market-wide volatility begets price instability and thus more volatility, whereas large single-stock price movements are more likely to lead to price (and volume) stability. In sum: volume reversions suggest price stability and persistently elevated volume suggests price uncertainty. We will test this hypothesis and look at where volume takes place, lit or dark, in our next piece in this series.

  • Philip Pearson

    CFA, Director, ITG Algorithms
  • Ben Polidore

    Managing Director, Head of Algorithmic Trading