Volatility of Volume

 

Stepping away from my normal weekly posts for a minute. I took some time to do a little research and thought I'd share my findings. All the data is sourced from either Refinitiv Eikon or DataStream. 

I recently saw some posts on Twitter about how much the top five names in the large indices impacted 2021 performance. I didn’t get a chance to look at them in detail. When I started to look into this myself, I found that while the contribution to the performance was rather high compared to an equally weighted portfolio, they weren’t outlandish. The data shows when there was smaller the number of the index constituents, say the NASDAQ 100, the larger the top names overperformed their weights. Below are some screenshots showing the top 5 names by index weight and their performance for the NASDAQ 100, S&P 500, Russell 1000, and the wider Russell 3000.

I couldn’t remember where I saw the original chart or what the point being made was, so I sort of gave up. Regardless of what the original article’s point was, this sent me down another road. As a former trader, I wondered if these ratios might also be visible in the volume traded. As I looked at volume, I realized that raw volume probably wasn’t the best metric to use. Lower priced stocks would have a huge outsized impact in shares traded though they might not be trading as much in Dollar terms. Let me just give you a frame of reference. I first looked at the S&P 500 constituents. The top five names by year end weight accounted for only 6.7% of total shares traded. That’s 37 billion shares against 555 billion for the index as a whole. Apple is the one common thread it traded both the most volume and had the most notional turnover, which is Price x Volume for each trade. The top five names by volume alone were Apple, Ford, Advanced Micro Devices, Bank of America, and AT&T. Each of those stocks traded north of 10 billion shares.

Next, I looked at index constituent turnover. I started with two of the smaller, by ticker count, indices. The NASDAQ 100 was nearly spot on. The top five names were 46% of the index performance and as the chart notes below, 45% of the total index turnover. The S&P 500 was actually even a closer match. The top five names were 22.7% of the 2021 performance and 22.8% of the index turnover. I was amazed by this, and I’d love to understand why this might have happened. I’d guess some of this has to do with the rise of passive and quantitative investing, but that’s another discussion. I asked myself, “So what?” Why would anyone care about this?

Instead of going on to the Russell indices here, I found my way down another path, something I thought might be slightly more interesting and useful for traders. I wondered how volatile volume was on a daily basis. During the last week of December much of the trading community was on a holiday break. The average S&P 500 constituent traded about 57% of its average daily volume (ADV) for the last year during that last week. For the Russell 3000 names, it was a bit higher. The average name traded about 70% of its average day, but the median here was closer to the S&P results at 59% of the ADV. The average in the RUA index was impacted by a few Bio Pharma names that traded north of 10x their normal volume on some news. As we’ll see below, the names in that Industry Group are quite volatile.

Everyone knows that the volume during that week between Christmas and the New Year is usually on the lighter side. This is nothing new, but have you ever looked at just how volatile the daily volume can be? I took a look at the most volatile names by volume using the year end Russell 3000 index constituents. Let me first explain what we’re going to look at in these tables below. Going from left to right, you first have some descriptive info. Starting with the RIC, which is Refinitiv’s ticker, the Issuer Name, year-end index portfolio weight, market capitalization in million USD, GICS Sector and Industry Group. The next section shows either volume or turnover in millions, but both have the same fields. Total traded, daily average, standard deviation, Coefficient of Variation, which for non-geeks is Standard Deviation divided by Average, then the max, minimum, and median for daily numbers. The Coefficient of Variation helps normalize the variation about the mean. The last set of data is the true value in these reports. This section shows the variation around the trailing 20 day average. Here we look at some of the same data points as the previous data set, but because the deviation can be a positive or negative here, I also include the absolute value numbers.

The first thing I wanted to look at were the top traded names. I took the top ten names traded by both volume and turnover. If you look at these next to one another, you will see some similar names. As mentioned above, Apple appears in both lists. It is both the second most traded by volume and total value. Meme stock, AMC was the top traded by volume, but it looks like it barely belongs in the top ten by turnover value, as it’s about 15% of the top turnover name Tesla. Even these mostly mega-cap names have a ton of variability when trading around their 20 day average. Every name in the top 10 has an absolute mean deviation from the 20 day trailing average of more than 20%. I’ll touch more on this metric below.

Next, I’m showing the same data fields, but the top 10 names by volatility of volume and turnover. For this rank, I’m using the coefficient of variation of the daily numbers. You’ll notice here that both lists are all BioPharma names. This isn’t all that unexpected. The Pharmaceuticals, Biotech, and Life Sciences Industry group has the largest number of names in the Russell 3000.

If we look at the index turnover on an Industry Group level, we can see even though Pharma, Biotech, and Life Sciences Industry Group accounts for 14% of the total names with 442 companies, it’s nowhere near the top in terms of market cap, where it ranks 21st of the 24 TRBC Industry Groups. It is the top according to volume, accounting for 12% of total volume. On the turnover side, it comes in 6th of 24 groups, accounting for 7% of index turnover. The standout data point in this view is Automobiles and Components, which is only 2.7% of index market cap, but is almost 7.5% of the total turnover. Mostly supported by Tesla’s outsized turnover, but Ford and GM are also in the top 100 turnover. Plus, Rivian’s late year explosiveness ranks it at 131 in total turnover in only two months of trading.

The final segmentation of the data I wanted to look at was the deviation from the 20 day average volume. Most traders use a 20 or 30 day historical look at volume to see how much a stock has been trading. This gives about a month of data depending on if you’re using 20 trading days or 30 calendar days, some systems vary on this. I’m looking at 20 trading days here. The numbers were a bit shocking to me. I’m sure more experienced quants could find fault in this methodology, but I’m not looking to be perfect here. I was just trying to see if the numbers would tell me anything. They did. While the upside can get quite large, the downside deviation is limited to -100%. That’s one limit to this. Taking that into account, I think this is still valuable. Let’s look at the top 10 names sorted by the average absolute values. This means if the volume was 20% lower than the average instead of -20%, it shows 20%. I’m just looking for the biggest differences. Many of these in the overall top 10 have not for more than a few months. Let’s look at the number two name, which has been trading for about five months. Adagio Therapeutics has one giant volume spike in mid-December, then trades about 90% less than the 20 day average for the remainder of the month.

Because small caps dominated the top 10, I decided to run a filter for large cap names. You can see that even in larger names there was a ton of volatility in the daily volume. To probably no one’s surprise, GME and AMC are high on this list. Seeing Clarivate, McAfee, Doordash, and Rivian on the list too made me realize that this list was dominated by the meme stocks. I then upped the filter to mega caps with a market cap over $100B. Here we have some rather large companies, and there’s still a large amount of volatility around the 20 day average. IBM was the biggest surprise in this list. It averages a 33% difference in either direction from the average. It also has a 23% median difference. Blackstone also appears. It caught my attention because it had a rather high max difference of 585% so for the next 20 days, the stock only traded about 50% of it’s 20 day average. Why is that important?

Many pre and post trade analytics that gauge risk and performance of putting a trade on use a standard trailing 20 or 30 day average volume in their inputs. I’m not arguing the best way to measure a trade’s impact, that’s another post. What I am saying is, if you use a trailing 20 day average volume as an input, you might be exposing yourself to some issues. Even if you have a more complex formula that gives more recent volume a higher weight and decays the weights of more historical data, there still an impact of using this somewhat stale historical data. Below is a screenshot of the volume for S&P 500 constituents on January 6th, it shows the volume at time versus a 20 day average. This is as of about 2:30pm. Looking at the top name here, you can see Humana (HUM) is trading about 15x the trailing 20 day average volume. If you were looking to trade 500,000 shares of HUM today, you’re more likely to have an easier go of it versus what you would have in the previous 20 days. This wouldn’t show up in too many pre-trade analytics.

I’m excited, because at Refinitiv, we’re working on a new Trade Performance Analytics (TPA) platform. As I mentioned, most trading cost analytics solutions are static or focused on post-trade reporting. Our analytics will provide a real-time agile platform that embeds standard analytics, like instrument-level statistics and execution benchmarking with the ability to integrate custom analytics for orders and trades. You can easily add custom analytics, which are open and extensible via Python. Right now, we’re focused on improving the framework for viewing this data. The next step from there will be incorporating intraday-dynamics of volume (like predicted volume), volatility and spread and predict those dynamics using machine learning. We’ll then use these intra-day predictors as featured inputs to refine our impact model. Below are some screenshots from a few of the windows in early versions of this platform.

Thanks for reading. If you have any questions about what we're working on, please give me a shout. Contacts below.

Twitter: @tanney9

LinkedIn: https://www.linkedin.com/in/michaeltsmithii/

Email: michael.smith2@lseg.com