high frequency trading: economics, empirics & politics bruno biais (toulouse school of...
TRANSCRIPT
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
NASDAQ
ARCA
NYSE
BATS
EDGX
EDGA
NASDAQ BX
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 talkDiscussion 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
Limit orders
Ask = 100,01Midquote: 100Bid = 99,99
New info:value = 100.02
HFT observes info quickly:Buy at 100,01
Slow observesinfo, wants to cancel bid: too late
t
Info eventually observed by all, midquote rises to 100,02
Fast Profit Slow loss
Implication
Ask = 100,01
M = 100
Bid = 99,99
HFT buys
t
Ask = 100,03
M = 100.2
Bid = 100,01
HFT marketable buy order followed by increase in midquote
Hendershott & Riordan (2013): Algorithmic trades’ impulse response
Ask = 100,01
Bid = 99.99
New info:value = 100.02
1st HFT observes info quickly: cancels limit order to sell at 100.01
Slowobserves Info: wants to cancel ask, too late
t
2nd HFT observes info quickly: buys at 100,01
Lower adverse selection cost for fast limit
Fast No Loss Fast Profit Slow loss
Baron, Brogaard, Kirilenko (2012)
CME E.mini S&P 500 futures
HFTs average profits: $ 0.77 per contract
Marketable orders: $ 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 roundtripNY-London: from 64.98to 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
“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.”
Hirshleifer (AER, 1971)
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
Excess equilibrium investment
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.”
Posner (NBER WP, 1974)
Privately optimal to be fast if
– C >
Equilibrium investment in fast trading techno(Biais, Foucault, Moinas, 2012)
Trading gain | fast
Investment cost
Trading gain | slow
Privately optimal to be fast if
– > C
My investment depends on the fraction of others’ that are fast ()
Trading gain | fast
Both decrease with :My marketable orders face larger spreadMy limit orders are more picked off
Which one decreases faster with ? or ?
Determines whether – increasing or decreasing
Trading gain | slow
– in Biais, Foucault, Moinas (2012)
0 1
Large : slow bear larger adverse selection costs until evicted
Small : fast suffer more from higher spread because trade more
C large => equilibrium investment = 0
C
0 1
.
For all advantage of being fast < cost
Small C => equilibrium investment =1
C .For all advantage of being fast > cost
Intermediate C: multiple equilibria
C
. . .
Interior equ: s.t. advantage of being fast = cost
For advantage of being fast > cost
Strategic complementarity
If increase in hurts slow more than fast
increasing in
Relative profitabiliy of investment increases with
Investment decisions are strategic complements
(Local as well as global complementarities)
Complementarity => equilibrium multiplicity
(but stable not **)
Contagion
If I expect all the others to be fast (I expect )
I must also be fast, lest I should be evicted
We all do the same: 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 1st 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 1st, 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)