Under normal circumstances, the cumulative value of the assets that comprise an ETF and the value of the ETF itself move and trade in lock-step. But August 24 was not a normal trading day, as there was a breakdown between ETFs and their corresponding intraday Net Asset Values. The question is: How should abnormally large differences in iNAV and ETF values be handled?
For capital markets participants, August 24, 2015, may have felt like post-traumatic stress.
With US markets opening down nearly 6%, there were unfortunate similarities to 2008, particularly the loss of order in the basic pricing of securities. One of the clearest cases of this was the breakdown between ETFs and their corresponding iNAVs (Intraday Net Asset Values).
As we all know, ETFs are priced off of a basket of components. Under normal circumstances, the cumulative value of the assets and the value of the ETF move and trade in lock-step, which makes sense because there are clear formulas that allow users to sum the ETF components and cash components to arrive at an ETF value in real time. ETF market makers ensure these values are in line nearly all of the time.
In this first of several articles for TabbFORUM, we explore how three ETFs functioned on August 24. Specifically, we plotted the difference between the ETF and its corresponding iNAV using the midpoint of quotes for the ETF and iNAV values. In the two subsequent notes, we plan to explore two areas: the number of halts available, and volume of trades that occurred during periods of large deviations. In the third note, we will focus on how much liquidity should be available in the ETF based on the bid/ask sizes of the components and vice versa.
What we measured
We examined three ETFs – SPY, XLF and XLK – plotting the ETF value minus iNAV. The iNAV values were computed with second resolution market data, while the plots below are the 60-second rolling average. The prices used for all computations were the mid-point between the bid and ask at each second. It’s important to note that quotes are indicative and don’t convey the same information as trades. As such, we explore the volume of trades that took place far outside of normal ETF minus iNAV bounds in a subsequent study. ETF components weightings and cash values were sourced from several sites on the Internet and verified with several market makers. Cash components are assumed to be constant for the day. We use August 17 as a reference because it was a recent “normal” day with low volatility.
SPY minus iNAV
SPY is interesting to examine because of the breadth and scale of its components. It shows the smallest divergence between iNAV and ETF value. Exhibit 1, below, is a plot of the difference between the SPY and the iNAV plotted on two days. The dark brown line is the average day, August 17; the light brown line is August 24.
On the 24, there were periods of time near the open when the SPY and its iNAV diverged by more than $1, or just under 1%, which is a huge divergence when compared to other days.
This is roughly two orders of magnitude greater than a typical day.
XLF minus iNAV
The situation becomes more interesting when looking at XLF, which has 90 components. Exhibit 2, below, shows substantially greater differences during the morning of the 24th. Using our rolling 60-second average around the open, there was as large as a 10% difference between the iNAV and the ETF. The individual data points show much larger maximum differences, as large as a 20.7% at 9:30:44.
The plot below shows something else of note: how long these differences remained. Within 20 minutes the difference returned to a more normal range. The spread between the two was elevated all day, which would be expected given the volatility of the day. The length of time these differences persisted is one of the more important questions around this issue, especially, how long should these differences be allowed to persist for and what size spread constitutes a problem?
This divergence is nearly three orders of magnitude larger than a normal day.
XLK has 76 components. Exhibit 3, below, shows the magnitude of the outliers near the open. The differences in XLK converged briefly then diverged and then returned to near average levels within 30 minutes. XLK presents a somewhat different behavior than the previous two ETFs in that its spread was somewhat higher for the duration of the day. Otherwise, its performance was similar to SPY and XLF in having a large difference for the 30 minutes after the open.
Implications and questions
The key question is how should abnormally large differences in iNAV and ETF values be handled? If the ETF trades substantially below its iNAV, should those trades be considered clearly erroneous? Or, what if a component trades well below the implied iNAV? If so, should they not occur in the first place? How many components of an ETF can be halted before the ETF can’t be priced? Should this cause the ETF to be halted?
I don’t believe that there are necessarily any clear answers to these questions. Different participants may have very different views. I’m eager to hear what others have to say.
As I stated above, this is the first of several notes I look forward to sharing through TabbFORUM to analyze:
Trades on ETF and components that happen substantially below the NAV of the components and or the implied component values.
Number of halts as it relates to each of the above differences.
Implied liquidity of ETF vs. components and vice versa.
Markets are highly interdependent, which is true for venues, products, technology, regulations and market makers.
It’s best if we all can find a way to work together.
It is reasonable to expect that some differences will happen, but how long should these differences be allowed to persist? Is executing “correctly” in this environment a broker’s, an exchange’s or market maker’s responsibility?
Sources: Tabb Forum