Editor’s note: this was originally published by KCG and is reprinted here with permission.
Market on Close (MOC) ETF trading can cause volume shocks that disconnect ETFs from net asset value (NAV). Illiquid and high beta ETFs seem most at risk of mispricing in MOC. This creates tracking errors that make ETF PMs look bad too.
Crib Sheet:
- ETFs depend on arbitrageurs to hold their price at fair value (NAV). And arbitrageurs depend on having a cheap and reliable hedge to offset risks.
- The close is one time that arbitrageurs can’t hedge. This makes market making in ETFs in MOC more risky.
- Data shows that disconnects from NAV in the MOC are common, but they are typically pretty small for most US ETFs, most of the time. This is good news for investors. It indicates that the markets are surprisingly efficient despite the lack of riskless arbitrage.
- However, disconnects do occur, mostly in less-liquid ETFs and those with higher beta, which may indicate that investors are paying more for MOC liquidity than they realize.
- Importantly, most benchmarks track index close, not ETF close. Consequently, we recommend investors with large MOC trades target NAV-close. This can be done via our ETF desk.
The importance of arbitrage for ETFs
In our recent report on intraday ETF trading we highlighted that 90% of US ETFs trade inside the spreads of their underlying baskets – effectively a no-arbitrage zone (exhibit 2). In addition, 42% of ETFs never traded in the arb-zone – accounting for 65% of ETF value traded.
But what about on the open (MOO) and close (MOC)? Well, that’s where things can get interesting.
ETFs typically trade very little MOC
Most stocks trade around 8% of their ADV on the market close (as we show in our September 2014 chartbook). However, ETFs tend to be more active in the morning – and many have a muted close. Exhibit 1 shows that most ETFs trade less than 2% of their ADV in the close.
There is no arb on the close
It’s good that large ETF trades typically avoid the close, because on the close it isn’t possible to arbitrage an ETF. Whenever ETFs are closed to arbitrage spreads tend to widen and the link bwtween the ETF and NAV weakens.
In the US close, both ETFs and underlying stocks close at the same time. Because of this it’s not possible for market makers to execute a stock hedge once they take on an ETF position. In fact, one of the only hedging tools available after 4pm is SPU futures, which trade until 4:15pm.
To account for this risk, market makers are less able to absorb large trades without moving ETF prices away from the last basket value. This can cause MOC to disconnect from NAV.
The MOC results are better than we expected
Despite all this, the average tracking of NAV by ETFs into the close is better than we had expected – especially given the volume shocks that we do see on the close (see appendix).
In many cases the average deviation is around the same as the intraday spread on the ETF (Exhibit 1 shows the majority of ETFs close within 5 bps of NAV on average).
Does the transparent US close help?
The US has a very transparent close. MOC orders need to be in the system 15 minutres early – and any stock with a large imbalance will be published to the whole market.
This gives traders a chance to pre-hedge a position and offset the close imbalance. Although this may explain the relatively small disconnects for most ETFs on the close, academic studies show this pre-hedging can also move the market, typically very efficiently. So it’s important not to confuse good tracking to the close with large trades being “free” from impact.
- If you want ETF prices that are close to NAV, avoid both the open and the close.
- The open is harder to arbitrage as spreads in the underlying are wide and some stocks may be gapping on overnight news
Avoiding the biggest disconnects
Although average disconnects from NAV look benign, nobody should want to be on the wrong end of a MOC mispricing, especially if it was caused by their own trade. Looking at our findings there are three factors that should minimize the chance of causing an outlier:
1. Avoid volume shocks
In theory, disconnects should happen when large trades impact the liquidity available on the close. However, thanks to the transparency of the US close, we found that unusually large close volumes were often not a statistically significant predictor of price disconnect (we discuss this further in the appendix).
However, the stocks with the largest average disconnects tend to be illiquid to start, with even more illiquid MOCs. In exhibit 3, we see a spike in average disconnect for ETFs that trade less than $100,000 on the close. Many of these are also small circles, indicating the MOC is small in notional as well as a percent of ADV.
- Many of the ETFs with the most deviations from NAV are levered ETFs.
- Levered ETFs are even harder for arbitrageurs to trade into the close because the ETF itself needs to trade to recalibrate the fund for the next day.
In reality, most of the 1600+ ETFs don’t regularly trade on the close – so it’s easy for a medium-sized trade in a small ETF to slip under the radar at a busy time like the close.
2. Beware high-beta & volatile ETFs
SPY with its beta of one is very easy to hedge, even after 4pm, via SPU futures. This helps keep its disconnect from NAV low despite reasonable MOC volumes. In contrast, higher beta and more volatile ETFs are harder to hedge. Exhibit 4 shows the clear relationship between historic ETF volatility and the frequency of large disconnects (over 10bps). The green color scale shows that Beta is also typically much higher for the stocks with more frequent large disconnects.
3. Why not trade the underlying instead?
Rather than risk pushing an ETF away from NAV, and the underlying index close, investors could execute a NAV-close trade (talk to the desk). A key twofold benefit of this is that you tap into the underlying basket liquidity at the time when stocks are at their most liquid. Both factors should make it cheaper to trade.
PMs can blame traders for bad TE
Typically the amount of risk you want an index PM to take is small, ideally zero. And tracking error (TE) is the most common metric to measure the amount of risk an index portfolio manager is taking.
But the ETF industry tends to use TE incorrectly, not recognizing two problems with their calculation:
1. You shouldn’t compare close-to-close ETF returns. To measure the risk of the ETF portfolio, you would calculate tracking error as the standard deviation of the differences in daily returns of the portfolio vs the benchmark.
Unfortunately, the industry tends to calculate TE using the ETF prices (not the portfolio). This means all the trading-related disconnects from NAV (that we discussed above) are included in the TE calculation.
2. Annualizing TE exaggerates trading disconnects. It makes sense to convert the daily return difference into an annual
number. That way it can be easily compared to per-annum returns and outperformance. Doing this, however, involves multiplying the standard daily deviation by the square root of time (√252). For example, the standard disconnect for SPY in Exhibit 1 is just 2.9 bps per day – around the same as the underlying basket spread. But this extrapolates to a tracking error of over 46 bps (exhibits 5 and 6).
The √ takes into account a normal amount of mean reversion over the rest of the year, but there is a natural pull to NAV in mispriced (overbought or sold) ETFs. For example, the autocorrelation of the daily disconnect for SPY is -46%. This strongly negative number shows that an overshoot is very likely to be corrected the next day.
Consequently, industry TEs don’t say much about the portfolio’s performance. In fact, TE is artificially high for ETFs.
- The tracking in ETFs is self-correcting.
- An ETF that closes rich because of a large buy trade typically underperforms next day, as arbitrageurs pull the ETF back to NAV.
A better way to assess ETF tracking
The issues above are key reasons we don’t like to use tracking error to measure ETFs or ETF managers. There are better ways. In fact, ETF.com utilizes a measure they call, “tracking difference”, to evaluate how well index-tracking ETFs work.
Using a metric like tracking difference is similar to what you can see when you eyeball longer-term data (exhibit 7a below). This is also much more useful when assessing the tracking for international ETFs who’s NAV close is at a different time to the US ETF’s close.
MOO is also more expensive to trade
At the open, stocks are all adjusting to the overnight news. Consequently, the underlying stocks tend to trade with more volatility and wider spreads – as we highlight in our September Chartbook. This adds to the risk and cost of market making in the ETF too. So it’s not surprising to see that ETF spreads are also wider in the early part of the day (exhibit 8).
In fact, in the first minute of trading, when some stocks may not have officially opened, we saw the median ETF spread at 33 bps. That’s almost six times wider than the spread at the end of the day.
Typically, wider spreads and more volatility make it more expensive to trade stocks and ETFs. That means that unless you’re one of the investors with a macro trade based on overnight news, you’ve got to be careful not to trade too aggressively in the first 30 minutes of the day and incur unnecessarily high trading costs.
APPENDIX: How did we do this?
We looked at just under one year of trading data, for ETFs with US underlying so NAV was in sync with the US Close. We also limited our sample to ETFs with consistent MOC trading – a total of 133 ETFs are in the sample.
We collected daily MOC volumes on a stock-by-stock basis and compared them to average MOC trading for that stock. We then looked at close prices of the NAV basket and the ETF, and computed the disconnects (rich or cheap) from NAV.
Finally, we compared the disconnects from NAV (in absolute value space) with the volume shocks on the close. This created charts similar to exhibit A, and a whole lot of metrics for each ETF, which we’ve used to compare ETFs in this report.
In general, the US ETF MOC is very efficient
For most ETFs, especially liquid ETFs, the market was able to digest unusually large MOC trades well.
The SPY chart (exhibit A1) shows that high MOC days in December 2013 caused disconnects around 5-6bps. Then another volume shock in April 2014 led to a difference of less than 4bps from NAV. We see similar patterns across most liquid ETFs.
Volume shock wasn’t a great predictor of price disconnect
Across the 133 ETFs that we studied, representing the more liquid US underlying ETFs, the volume shocks proved to be a poor predictor of price shocks (very low R2). In fact, only 3% had R2 over 20%.
Even for illiquid ETFs like IYF (exhibit A2), which had some very extreme MOC volume shocks, the correlation with price disconnects was far from strong, at 26%.
Beware of sector, mid cap and style ETFs
Seven of the 12 strongest R2 came from sector funds like XBI (exhibit A3). We found that of the 15 strongest R2’s:
- Sector ETFs had the most extreme MOC volume shocks.
- 5 were style funds.
- 2 were mid-cap ETFs.
Often we found that the normal MOC volumes in these examples were especially low – making their volume shocks that much more significant. Intuitively, this may be because they are used a lot by hedge funds who short them – making their trading is less concentrated into the close.
