Björn Hagströmer

Professor of Finance at Stockholm Business School, Stockholm University, doing research on financial market microstructure, with applications to asset pricing, financial econometrics, and liquidity measurement.

Photo by Maria Stoetzer

Robothandel och volatilitet

Kan volatiliteten på börsen nu under Coronakrisen kopplas till robothandeln? Det frågade Rapport igår och SvD idag. Här kommer lite bakgrund till det jag säger i intervjuerna.

blinkning

Jag och Lars Nordén har undersökt hur robothandel, eller högfrekvenshandel/HFT som vi forskare hellre säger, påverkar volatiliteten på den svenska marknaden. Vi visar att HFT framförallt agerar som market makers, och att de genom sin aktivitet dämpar prissvängningarna. Undersökningen finns här. Det ska dock betonas att den är gjord på data från augusti 2011 och februari 2012. Augusti 2011 var en ganska volatil månad, men inget i närheten av det vi ser nu.

En annan studie, gjord av Jonathan Brogaard m fl på amerikansk data från 2009-2010 (här), bekräftar bilden att HFT har en dämpande effekt på volatiliteten i enskilda aktier. När hela marknaden är volatil, däremot, visar de att HFT byter strategi och i viss mån kan driva på volatiliteten. I en studie av Flash-kraschen i maj 2010 (här) kommer Andrei Kirilenko m fl till en liknande slutsats. Deras undersökning visar att HFT inte kan lastas för kraschen, men att deras agerande under den dagen kan ha förstärkt volatiliteten.

Jag hänvisar i intervjuerna också till att HFT tjänar pengar på volatiliteten. HFT är bra på att förutse prisförändringar på kort sikt. Det har visats i flera artiklar, och jag och mina kollegor har också sett detta på den svenska marknaden (här och här). Direkt bevis för att de tjänar pengar nu under Coronakrisen får man om man tittar på några av HFT-firmornas börskurser. HFT-firman Virtu är upp 37% sen 3 feburari, och Flow Traders är upp 29% under samma period.

Stäng inte börsen

Jag skriver på DI Debatt (20 mars) att förslag om blankningsförbud och börsstängning som kommer nu i Covid19-krisen bör avvisas. Här lite extramaterial, som inte fick plats i tidningen.

Bakgrund

  • Blankningsförbud har den senaste veckan införts i Sydkorea, Spanien, Frankrike, Italien och Belgien (BloombergFT)
  • Jag har inte sett att någon börs stängt helt, men NYSE och London Metals Exchange har aviserat att de migrerar sin golvhandel till elektroniska system.

Empiriska analyser av finanskrisens blankningsförbud (bakom betalväggar, men det går säkert att hitta gratisversioner på SSRN)

Andra inlägg om stängning av börsen

  • Daniel Kederstedt väckte frågan om att stänga börsen i SvD Näringsliv, 13 mars.
  • Stockholmsbörsen har stängt ett par gånger i historien, men då får man gå tillbaka ända till 1930-talet. För en genomgång av dessa fall, se Daniel Waldenströms inlägg på Ekonomistas, 17 mars.
  • Susanna Sällström Matthews förespråkar stängning av börsen på DI Debatt, 20 mars.

Auction volatility safeguards

At the opening of US equity markets on August 24, 2015, prices dropped sharply following overnight turmoil in global markets. In the minutes after opening, large order imbalances triggered circuit breakers to halt trading in numerous stocks and exchange-traded funds. More than a thousand securities were affected and hundreds of them experienced repeated trading halts before prices stabilized.

Large price moves at the market opening is nothing new. To effectively incorporate overnight news in prices, equity markets typically open with a batch auction that aggregates trading interests and establishes an equilibrium price. What caused chaos on August 24, 2015, is that many opening auctions apply price collars, disallowing extraordinary price moves in the auction mechanism. The consequence is that large imbalances spillover to the continuous trading, where continued price swings trigger circuit breakers. In response to this problem, US exchanges—including NASDAQ, NYSE, and NYSE Arca—have suggested replacing auction price collars with another type of volatility curb: a volatility extension.

In a new paper that I coauthor with Ester Félez Viñas, we analyze the merit of Call Auction Volatility Extensions (CAVEs). The mechanism works as follows. If there is a large price change during the batch auction, the order entry phase (the batching period) is extended to allow investors to reconsider their orders. Instead of constraining price discovery (like price collars do), the volatility extension delays it.

Volatility extension

We study the introduction of CAVEs in the closing auction at NASDAQ Stockholm. On December 1, 2014, NASDAQ introduced CAVEs for its Stockholm segment, while the Copenhagen and Helsinki segments were unaffected. The introduction thus provides us with a nice benchmark.

The figure below shows the incidence of extraordinary volatility events before and after the introduction of CAVEs. It is evident that the extraordinary volatility incidence drops substantially in the Treatment market (NASDAQ Stockholm), in particular for small-caps. The finding is corroborated by that the Control markets (NASDAQ Copenhagen and Helsinki), where no CAVEs are introduced, do not show the same effect.

Fig1

In the paper we also analyze why extraordinary volatility is reduced. We hypothesize  that CAVEs reduce the risk of market abuse and that it increases investors’ trust in the auction mechanism, and present supportive empirical evidence.

The paper is available at SSRN.

Celebrating Campbell-Lo-MacKinlay

On May 4-5, together with Albert Menkveld I organized the Conference on the Econometrics of Financial Markets at Stockholm Business School. The conference celebrated the 20th anniversary of an influential book on financial econometrics, published in 1997 by John Campbell, Andrew Lo, and Craig MacKinlay.

IMG_5760The conference featured keynote speeches by John Campbell and Andrew Lo, as well as presentations of academic papers at the research frontier of financial markets. The topics covered included the automatization of financial markets, systemically important financial institutions, network effects in financial markets, and the intriguing question whether finance can cure cancer.

More info about the event, including the program, is here: sbs.su.se/EFM2017

Overestimated effective spreads

Accurate liquidity measurement is important for liquidity timing and order routing. One of the most prevalent measures is the effective spread, defined as the percentage difference between the transaction price and the bid-ask spread midpoint. For example, if the quotes for a stock are $10.00 to sell and $10.01 to buy, the effective spread of a buy trade at $10.01 is half a cent (about 5 basis points).

To see the logic of the effective spread, consider the market maker who sells to the incoming trader. The market maker provides the service of immediate execution, and in return earns a premium relative the fundamental value of the shares. The effective spread captures that premium, using the spread midpoint as a proxy for the fundamental value.

In a new paper, I challenge the use of the spread midpoint when measuring the effective spread. I show that the use of the midpoint leads to an overestimation of the “true” effective spread.

Assume that the fundamental value of the stock in the example above is 10 dollars and 0.25 cents. The effective spread is then asymmetric; 0.25 cents for trades on the bid side and 0.75 cents (three times higher) on the ask side. If traders care about transaction costs, the relatively wide ask-side spread deters buyers, whereas the tight bid-side spread may attract sellers. There are then more traders submitting market orders at the bid side, and the true effective spread is, on average, smaller than the average midpoint effective spread (which is 0.5 cents).

How bad is it? I analyze a sample of five trading days for the S&P500 stocks and find that the effective spread is overestimated by 16% on average. The problem is worse in stocks where the tick size is high relative the trade price. For example, the picture below shows that the effective spreads of stocks priced between $5 and $15 are overestimated by 60%. Stocks priced higher than $100 are unaffected by the bias.

Can the overestimation problem be fixed? I propose that a viable alternative is to measure the effective spread relative the microprice instead of the midpoint. The microprice is a quote-volume-weighted average of the best bid and ask prices, and it is a commonly used proxy for the fundamental value.

Should investors care? Trading in the US stock market is highly fragmented. As an aid to investors struggling to make their order routing decisions, Rule 605 of RegNMS mandates all exchanges to publish their average effective spreads for each stock on a monthly basis. I show that the overestimation influences the ranking of venues. For example, for stocks priced lower than $25, NYSE MKT (abbreviated “ASE” in the graph below) has the lowest ranking when the effective spread is measured relative the midpoint (left side of the graph below). When I instead measure the spread relative the micoprice, NYSE MKT is ranked highest of all the venues! (see the right side in the graph below) The conclusion is that the midpoint effective spread is potentially misleading the order routing decision.

The article in full text is available at SSRN.

A network map of information percolation

When the trading of a security is fragmented across exchanges, MTFs, and dark pools, the prices at different markets co-move almost instantaneously. Almost. What happens in the time span specified as almost instantaneous? Once new information appears at one market, how does it spread to the other markets trading the same security?

Information percolation describes how information spreads between markets. In a paper with Albert Menkveld, we propose a new methodology to measure information percolation in ultra high-frequency data, uncovering the paths information flows take between markets. We apply it to the euro-Swiss franc currency pair, showing how information flows between dealers at the OTC market and the interdealer trading platform EBS.

The paper is at SSRN, and a longer blog post is at Albert’s website.

FX_2015_100ms_inf_period3_community_new_2