High Frequency Trading Economics Empirics Politics Bruno

High Frequency Trading: Economics, Empirics & Politics Bruno Biais (Toulouse School of Economics, FBF IDEI Chair on Investment Banking and Financial Markets) Presentation prepared for the Banque de France Conference on Algorithmic and High-Frequency Trading November 2013

Lots of information in financial markets Difficult and costly to process Slower than the others = too late

Fragmented markets Due to regulatory push for competition & technology Lots of quotes in different markets: Disappear quickly Hard to identify best trading opportunities before vanish

Financial institutions’ response: Investment in market technology Fast connection (fiber-optic, colocation, throughput) Clever terse code Computers that read Bloomberg faster than eye blinks Minimize latency = delay between info event & trade execution

This talk Discussion based in particular on the following papers: Biais & Foucault (2013) survey forthcoming in Bankers Investors & Markets Baron, Brogaard, Kirilenko (2012) [CME data] Biais, Foucault, Moinas (2012) [theory paper] Hendershott & Riordan (2013) forthcoming JFQA 1) Consequences of HFT, for given fast trading technology 2) Equilibrium investment in fast trading technology 3) Policy and politics

1) Consequences of HFT, for given fast trading technology 2) Equilibrium investment in fast trading technology 3) Policy and politics

Info. advantage for fast marketable orders, adverse selection cost for slow limit orders New info: value = 100. 02 Limit orders Ask = 100, 01 Midquote: 100 Bid = 99, 99 HFT observes info quickly: Buy at 100, 01 Slow observes info, wants to cancel bid: too late Info eventually observed by all, midquote rises to 100, 02 t Fast Profit Slow loss

Implication Ask = 100, 03 M = 100. 2 Ask = 100, 01 M = 100 HFT buys Bid = 100, 01 Bid = 99, 99 HFT marketable buy order followed by increase in midquote t

Hendershott & Riordan (2013): Algorithmic trades’ impulse response

Lower adverse selection cost for fast limit New info: value = 100. 02 Ask = 100, 01 Bid = 99. 99 1 st HFT observes info quickly: cancels limit order to sell at 100. 01 2 nd HFT observes info quickly: buys at 100, 01 Slow observes Info: wants to cancel ask, too late t Fast No Loss Fast Profit Slow loss

Baron, Brogaard, Kirilenko (2012) CME E. mini S&P 500 futures HFTs average profits: Marketable orders: $ 0. 77 per contract $ 0. 93 per contract (at the expense of institutional traders) Limit orders: $ 0. 33 per contract (losing money versus HFT and market makers) Teaser Biais, Declerck, Moinas (in a few minutes)

Industrial Organization of Liquidity Supply Fast limit less exposed to adverse selection than slow limit Two countervailing effects for liquidity supply: 1) Lower cost for fast liquidity suppliers: makes liquidity cheaper 2) Slow limit orders, exposed to winner’s curse, exit market => Fast face less competition from slow, better able to extract oligopoly rents (// Biais, Martimort, Rochet, Econometrica 2000): makes liquidity more expensive Which effect will dominate? Ambiguous.

Benefits of HFT for society Reduces adverse selection cost for liquidity supplier: tends to increase quantity of liquidity supplied => larger depth at quotes and to reduce cost of liquidity supply => tighter spread Facilitates arbitrage across markets & helps linking fragmented markets

Costs of HFT for society Fast marketable orders => adverse selection costs for limit orders Deters placement of limit orders by slow traders => reduces competition to supply liquidity Tends to reduce quantity of liquidity supplied => lower depth at quotes and to increase cost of liquidity supply => larger spread

1) Consequences of HFT, for given investment in fast trading technology 2) Equilibrium investment in fast trading technology 3) Policy implications

Social costs and private gains Some of the social benefits of HFT are aligned with private gains of high frequency traders: reduced cost of liquidity supply better alignment of markets But some of the social costs of HFT also are aligned with private gains of high frequency traders adverse selection costs of orders picked off by fast traders are mirror image of trading profits of fast traders

An example of the divorce between private and social efficiency Project Express: fiber-optic cable across Atlantic Reduces data roundtrip NY-London: from 64. 98 to 59. 6 milliseconds Handful of trading firms Subscribed Cost = $ 300 million Profitable for subscribers (otherwise stay out), not for society: cost of socially useless investment passed to slow traders

Hirshleifer (AER, 1971) “foreknowledge: whatever does actually occur will, in due time, be evident to all” “the distributive aspect of access to superior information … provides a motivation for the acquisition of private information that is quite apart from any social usefulness of that information. ” “There is an incentive for individuals to expend resources in a socially wasteful way in the generation of such information. ”

Excess equilibrium investment Investment in fast trading technology iff Private gain > private cost Since private gain includes profits from picked off limit orders, we can have Private gain > social gain Since private cost does not reflect adverse selection cost borne by slow traders, we can have Social cost > private cost Hence we can have Equilibrium investment > socially optimal investment

Posner (NBER WP, 1974) To the extent that investment in fast trading technology motivated by desire to earn rents Informational rents mirror image of adverse selection costs Market power rents if slow liquidity suppliers exit Opportunity cost of resources allocated to HFT “The existence of an opportunity to obtain monopoly profits will attract resources into efforts to obtain monopolies, and the opportunity costs of those resources are social costs of monopoly too. ”

Equilibrium investment in fast trading techno (Biais, Foucault, Moinas, 2012) Privately optimal to be fast if f–C>y Trading gain | fast Trading gain | slow Investment cost

$My investment depends on the fraction of others’ that are fast (a) Privately optimal$

My investment depends on the fraction of others’ that are fast (a) Privately optimal to be fast if f(a) – y(a) > C Trading gain | fast Trading gain | slow Both decrease with a : My marketable orders face larger spread My limit orders are more picked off Which one decreases faster with a ? f or y ? Determines whether f(a) – y(a) increasing or decreasing

f – y in Biais, Foucault, Moinas (2012) f-y 0 Small a : fast suffer more from higher spread because trade more 1 Large a : slow bear larger adverse selection costs until evicted a

C large => equilibrium investment a = 0 C . f-y a 0 For all a advantage of being fast < cost 1

Small C => equilibrium investment a =1 f-y C For all a advantage of being fast > cost . a

Intermediate C: multiple equilibria C . . . f-y a a* Interior equ: a s. t. advantage of being fast = cost a** 1 For a =1 advantage of being fast > cost

Strategic complementarity If increase in a hurts slow more than fast f - y increasing in a Relative profitabiliy of investment increases with a Investment decisions are strategic complements (Local as well as global complementarities) Complementarity => equilibrium multiplicity (but a = 1 stable not a**)

Contagion If I expect all the others to be fast (I expect a = 1) I must also be fast, lest I should be evicted We all do the same: a = 1 as expected => investment waves in HFT

1) Consequences of HFT, for given investment in fast trading technology 2) Equilibrium investment in fast trading technology 3) Politics & policy

If there is too much HFT, should we tax it? Theoretical misgivings: If we tax high message traffic, maybe we’ll deter the “good” kind of HFT (limit orders need to be modified or cancelled often to avoid adverse selection) without affecting the “bad” kind (adversely hitting limit orders) Practical awkwardness: French tax (August 1 st 2012) only for French firms, not foreign ones … Teaser J. E. Colliard this afternoon

Market response HFT free platforms Or give slow traders option to execute only against slow orders If slow traders find execution against fast costly, they will choose this option

Concerns If it is expected that there will be no liquidity on HFT free platform, nobody will go there, confirming expectation: “bad equilibrium” If HFT can influence exchanges’ policy, they could prevent exchanges from offering option to place HFT free orders (i. e. , orders precluding execution against fast)

An interesting experience Baron, Brogaard, Kirilenko (2012) quoted above Results pretty damning for HFT CFTC study, based on CME data CME sued by HFT firms: data supposed to be used for monitoring and regulation, not research ! Baron, Brogaard & Kirilenko had to withraw their study ! Officially, does not exist any more (but I still have it ; -) Suggests significant lobbying power of HFT firms

Systemic risk concerns HFT firms have very little capital HFT transactions occur at much higher frequency (many per second) than clearing (daily) HFT trade a lot with one another, often in same direction, or transferring hot potatoes If one HFT, with little capital, takes big loss => could go bust This could propagate to other HFTs => lots of uncleared trades, mess

Mitigating systemic risk Capital requirements for HFT firms Capital reduces risk of going bust Also increases skin in the game or fund manager, reducing temptation to gamble Stress tests, at level of each trading firm, to evaluate impact of shocks

Conduct pilot experiments Similar to pilot introduction of TRACE in US For example, instead of introducing two taxes on August 1 st, 2012 (one on daily trades, the other on HFT) It would have been a good idea to introduce the two taxes at different point in time (so that the effect of each one could have been evaluated independantly) and for a randomly selected pilot sample (to conduct diff in diff evaluation)