The Treasury Market Flash Event of February 25, 2021

Introduction

On February 25, 2021, there was an unusual “flash” event in U.S. Treasury markets. The prices of Treasury securities dropped sharply amid strained liquidity conditions, before recovering within about an hour. This event follows several similar episodes in recent years, including the flash event of October 2014, which was discussed in the Joint Staff Report on the U.S. Treasury Market on October 15, 2014 (PDF). Given the key role played by the Treasury market as the largest and most liquid sovereign bond market in the world, it is important for policymakers to understand the nature and causes of these episodes. This Note discusses what happened on February 25, what we currently know about the likely causes, and how it compares with previous similar episodes.

Leading up to the February 2021 episode, liquidity conditions in the Treasury market had largely recovered from the particularly severe episode of illiquidity that followed the onset of the COVID-19 pandemic in March 2020. The yields on Treasury securities had risen substantially from their pandemic-related troughs, in part following news of the arrival of COVID-19 vaccines and expectations of further fiscal support, as well as an increase in the issuance of longer-term Treasury securities. Since the beginning of 2021, the increase in yields became more rapid and market-implied measures of interest rate uncertainty had picked up; at the same time the volume of trading in Treasury securities rose notably.

On February 25 itself, yields had continued their recent upward trend through the morning. Shortly after 1 p.m. yields spiked sharply higher, which market commentary linked to the release of the results of a 7-year Treasury note auction at 1 p.m. We show that trading conditions in the most liquid parts of the Treasury market deteriorated notably around the same time, with signs that liquidity provision by high-frequency market makers was insufficient to meet the heavy order flow. We further show that the episode was much less severe and shorter-lasting than the episode of March 2020, with the cost of trading as measured by bid-ask spreads quickly returning to normal levels. However, as of early April 2021, the total size of posted quotes on electronic order books (“market depth”) has not yet fully recovered, which suggests that Treasury market liquidity may remain fragile even amid sizeable ongoing Treasury purchases by the Federal Reserve.

What Happened on February 25?

The prices of Treasury securities (which move inversely with yields) fell through the morning of February 25, amid elevated trading volumes, before dropping sharply lower shortly after 1 p.m., as shown by the black lines in Figure 1. Similar patterns were observed in prices in the Treasury futures market (not shown). The sharp spike mostly reversed within an hour, similar to some previous flash events in the Treasury market, including the flash event of October 15, 2014.

Market commentary pointed to the proximate trigger for the spike down in prices being the release of the results of a 7-year note auction at 1 p.m. The auction tail (the difference between the average and lowest accepted prices) was unusually large, which investors may have interpreted as an indication that the market price of Treasury securities would need to fall further. That in turn may have triggered strong selling pressure; indeed, the spike down in the prices was accompanied by a sharp increase in trading volumes for most tenors (the blue bars in Figure 1 show volumes for the cash market). While the downward spike in prices quickly reversed, prices nevertheless resumed their downward trend over the day.

While the increase in trading volumes for most tenors after 1 p.m. was notable, it was hardly unprecedented; and similar spikes in volume have not usually resulted in large flash moves in prices (for example, elevated volumes were observed for some tenors earlier in the day on February 25, when seller-initiated trades persistently exceeded buyer-initiated trades). So the flash move after 1 p.m. suggests that the increase in volumes was also accompanied by an unusual deterioration in liquidity conditions. In the remainder of this section, we confirm that conjecture by taking a closer look at quoting and trading activity.

Trading in the most liquid parts of the Treasury cash market and in the Treasury futures market takes place over a “central limit order book.” Market makers provide liquidity by posting limit orders—quotes to buy or sell a specified quantity of securities at a specified price—on a centralized order book. Other market participants can also provide liquidity by submitting limit orders or consume liquidity by submitting market orders, that is, instructions to buy or sell a given quantity of securities at the best available price on the order book. The total dollar amount of quotes at the best few prices (or “levels”), commonly referred to as “market depth,” is often considered to be a measure of available liquidity; the more quotes are posted on the top levels of the order book, the easier it may be to buy or sell a given quantity of securities in a short period of time without moving the market price materially.

Panels A and B in Figure 2 show cash market depth at the top one and three levels of the order book for the 10-year note and 30-year bond, respectively. Quotes to buy are shown in blue above the horizontal axis, while quotes to sell are shown in red below the horizontal axis. In both cases, the data are plotted at a millisecond frequency and show the maximum depth available in each millisecond. Depth dropped sharply shortly after 1 p.m. pointing to a reduction in the willingness of market makers to post quotes.

However, while market depth is commonly viewed as a measure of liquidity, in some past episodes when depth has fallen sharply, the costs of trading as measured by the bid-ask spreads between the best available buy and sell quotes have not increased markedly. Even if market makers reduce the amount of quotes posted to the order book, if they continue to replenish quotes rapidly in response to incoming orders, and if orders are broken into sufficiently small chunks, then there may be little observable impact on trading costs. This appears to have been the case for tenors up to 10 years on February 25. To see this for the 10-year note, consider the black dots in the upper panel of Figure 2.A, which show the total size of all executed orders to buy (above the axis) and sell (below the axis) the 10-year note. The total size of executed orders generally remained well inside the posted depth after 1 p.m., so bid-ask spreads (shown in the lower-left panel) were fairly stable.

However, for longer tenors (the upper panel of Figure 2.B shows the same measures for the 30-year bond) the total size of executed orders often exceeded depth at the top few levels of the order book, meaning that some transactions were taking place at relatively unfavorable prices, so measured bid-ask spreads (the lower panel) repeatedly widened temporarily, until the order book was replenished, when spreads tightened again. Thus, the volatility of spreads increased, as well as the average level. As discussed by Dobrev and Meldrum (2020), a coincident drop in market depth and increases in the level and volatility of spreads indicates that the speed of order book replenishment was insufficient to meet incoming order flow without leading to higher trading costs. That in turn suggests a reduction in the activity of market makers that post quotes at high frequencies using automated algorithms. Those high-frequency market makers likely include so-called principal trading firms, but also other market participants, including bank dealers as well as some hedge funds.

Why Did the February 2021 Flash Event Happen?

While it seems fairly clear that the Treasury auction on February 25 was the proximate trigger, the reasons that this caused selling pressure and a withdrawal of high-speed liquidity provision are hard to piece together with certainty. This is in part because we cannot deduce from the available data how automated trading algorithms work, what data they take as inputs, and when market participants might turn particular algorithms on and off.

However, a couple of factors are likely relevant. First, the February 25 episode came against a backdrop of rising yields (and therefore falling prices) on longer-term Treasury securities and greater uncertainty about future yields amid heightened trading volumes. The pace with which longer-term yields (shown in Figure 3) were increasing had been particularly rapid over the previous month, and there had also been a notable increase in forward-looking measures of interest rate volatility (shown in Figure 4), particularly for the 5-year rate. This backdrop seems likely to have been an important factor that influenced both how market makers managed their risk exposures and how market participants interpreted the signal about future yields from the February 25 auction results.

Second, high-frequency market makers may have felt that their algorithms were temporarily less reliable at determining the “correct” price. In general, high volatility need not be a bad thing for high-speed algorithmic market makers; if they can react more quickly to incoming data than other market participants, a period of high volatility may even present additional profit opportunities. However, that relies on their algorithms continuing to reliably determine the correct price even in turbulent market conditions, which in turn relies on stable relationships between different prices. While we do not know the details of the algorithms used by market participants, market commentary suggests that a breakdown in correlations between asset prices was a factor that caused some market participants to step back from posting quotes.

Comparison to Past Similar Events

Figure 5 puts the February 25 episode into context by comparing it with the episode of market dysfunction that followed the onset of the COVID-19 pandemic in March 2020. Panel A shows daily time series of market depth, daily average bid-ask spread and intraday volatility of the bid-ask spread for the 2-year (the green dotted line), the 5-year (the red line), and the 7-year notes (the blue dashed line), while Panel B shows the same measures for the 10-year note (the green dotted line), the 20-year bond (the red line), and the 30-year bond (the blue dashed line). Market depth declined sharply for all maturities on February 25, reminiscent of the declines observed following the onset of the COVID-19 pandemic in March 2020. However, the severity of the impact on spreads in February 2021 was far less than that in March 2020, particularly for longer maturities, while the subsequent recoveries in spreads have been much faster.

While market depth has largely recovered for most maturities since the February 2021 flash event, it remains relatively low for the 5-year note. This is an unusual pattern, because depth tends to recover quickly for shorter maturities following episodes of illiquidity. However, it seems consistent with the relatively large recent increase in uncertainty around the 5-year horizon, as discussed in the previous section. Of note, the experience of March 2020 suggests that while depth remains relatively low, the risk of an increase in trading costs may be elevated; the sharp spike in the level and volatility of bid-ask spreads at that time was preceded by a similar period of low market depth during which market functioning was more heavily reliant on high-speed liquidity replenishment.

We next put the duration of the episode into broader context by comparing it with nine other episodes (listed in Table 1) when depth in the Treasury market has dropped notably and suddenly during the past several years. For these purposes, we focus on the time it takes for depth to recover after the episode. Figure 6 shows the evolution of depth for the 10-year Treasury note on the days around the ten events. To assist in illustrating the speed of recovery, Table 1 reports the number of business days it took for a 3-day moving average of depth to recover at least 75 percent of the initial decline (defined as the difference between the average depth over three days preceding the episode and depth on the start day of the episode). We can separate the events into three categories:

• The left panel of Figure 6 and the upper panel of Table 1 show events with a quick depth recovery within a few days. Two of these three events were associated with other flash moves in prices—the equity flash crash of May 2010 and the Treasury flash rally in October 2014—when sharp movements in asset prices that were not associated with a clear fundamental trigger reversed quickly. The third followed the Brexit referendum, which delivered a surprise result but did not have a major impact on the uncertainty around Treasury yields.

• The center panel of Figure 6 and the center panel of Table 1 show events with a medium-paced recovery within a month or so. These events seem to be characterized by more substantial economic and financial market uncertainty, and include the June 2013 “taper tantrum” in the Treasury market, the February 2018 “VIX turmoil” when the volatility of stock prices rose sharply, and the episode of growing concerns among investors about the global growth outlook in August 2019. Based on the trajectory of depth recovery, the February 2021 episode appears to belong to this category, which appears consistent with the recent increase in uncertainty about longer-term interest rates.

• The right panel of Figure 6 and the lower panel of Table 1 show events with an even slower depth recovery taking months. These events had a more severe impact on financial markets, and include the Global Financial Crisis (dated here to coincide with the bankruptcy of Lehman Brothers in September 2008), the episode of investor concern about European sovereign debt positions in August 2011, and the market turmoil following the onset of the COVID-19 pandemic in March 2020.

In summary, although the available sample of episodes of notable sudden drops in market depth is small, and the above characterization is somewhat subjective, it does appear that the amount of economic uncertainty accompanying each episode has a bearing on the time it takes for market depth to recover. That observation raises the question of whether there is a systematic relationship between the recovery time and some observed measure of uncertainty. In particular, standard market microstructure theories emphasize the negative relationship between the realized volatility of asset prices and market depth (see, for example Nguyen, Engle, Fleming and Ghysels (2020)). However, Figure 7 shows that there is little correlation between the level of the realized volatility of the 10-year Treasury yield at the start of the episode and the time taken for market depth for the 10-year Treasury to recover to 75 percent of its pre-episode level. Unreported results also show there is little correlation between the same measure of volatility and the initial decline in depth. Thus, it appears that volatility may not by itself be a particularly good predictor of episodes of Treasury market illiquidity, or their severity and persistence. Furthermore, the time it takes for liquidity to recover may also depend in part also on actions taken by policy makers.

Although market depth has taken over a month to partially recover following the February 2021 episode, with longer-lasting impacts for short-and-medium dated securities than long-dated securities, measures of the cost of trading such as the level and volatility of bid-ask spreads subsided back to their prior levels within a few days (as they did following most of the episodes discussed above). This suggests that there was only a fairly short period where the velocity of order book replenishment by high-speed market makers was not always sufficient to meet order flow without moving prices and quoted spreads materially. However, the fact that market depth generally takes longer than spreads to fully recover suggests that liquidity may remain fragile for a period after such episodes, because liquidity provision appears to be more heavily reliant on high-speed replenishment by market makers to meet trading demand.

The Impact on Futures Markets

While the above analysis has focused on the cash part of the Treasury market where on-the-run securities are traded for spot delivery, activity in that part of the market is tightly linked to the futures part of the market where Treasury securities are traded for future delivery. As documented by the Joint Staff Report on the U.S. Treasury Market on October 15, 2014 (PDF) and Dobrev and Schaumburg (2015a, 2015b and 2018), there is substantial high-speed cross-market trading between cash and futures markets, including by participants who attempt to take advantage of relative mispricing between the markets. That raises the question of whether the deterioration in cash market liquidity also spilled over to the futures market.

Figure 8 shows daily time series of market depth, average bid-ask spreads, and the volatility of bid-ask spreads for three different Treasury futures contracts traded on the Chicago Mercantile Exchange. While depth also dropped notably in the futures market, the impact on bid-ask spreads was negligible, suggesting that order book replenishment remained sufficiently fast such that there was no material increase in trading costs. The finding that the futures market was less affected than the cash market is consistent with what we observed in March 2020, as documented previously by Dobrev and Meldrum (2020). However, in contrast to the systemic shock to liquidity in March 2020, unreported results show that the impact of the February 2021 event on other futures markets, including those for key equity indices and foreign exchange rates, was also limited.

Conclusion

The Treasury market flash event of February 25, 2021 underscores the pivotal role of high-speed liquidity provision in the most liquid electronic parts of the Treasury market. We find evidence that the sharp drop in prices that day was accompanied by a sudden drop in market depth and a brief deterioration in high-speed liquidity provision amid elevated transaction volumes, albeit to a much lesser extent than during the episode of severe illiquidity in March 2020. Similar to some previous episodes accompanied by moderately elevated economic and financial market uncertainty, market depth has recovered steadily since February 25 at a pace comparable to that observed following other such episodes, while high-speed liquidity provision appears to have rebounded fairly quickly. That said, market depth has taken over a month to partially recover, which suggests that Treasury market liquidity has been more heavily reliant on high-speed replenishment to meet trading demand and may remain fragile.

---

Disclaimer: FEDS Notes are articles in which Board staff offer their own views and present analysis on a range of topics in economics and finance. These articles are shorter and less technically oriented than FEDS Working Papers and IFDP papers.