LIQUIDITY, MARKET IMPACT, HFT : THE COMPLEX ECOLOGY OF FINANCIAL MARKETS Jean-Philippe Bouchaud, with: B. Toth, M. Wyart, J. Kockelkoren, M.

Transcription

1 LIQUIDITY, MARKET IMPACT, HFT : THE COMPLEX ECOLOGY OF FINANCIAL MARKETS Jean-Philippe Bouchaud, with: B. Toth, M. Wyart, J. Kockelkoren, M. Potters,

2 2 But is this that obvious? How does it work really? At any instant there is no reason why supply and demand should match Without a liquidity buffer, prices would respond by making huge swings Tâtonnement and price formation: Efficient market theory tells us that any up- or down-tick away from the «fundamental» price should attract sellers/buyers so that markets should function and be oh really? Market operators have long realized the need of intermediaries: Market Makers/liquidity providers

3 Market Impact: orders to buy/sell, even uniformed or random, impact prices up/down This may create more orders in both directions and cascade: An important piece of information is the price itself selfreflexivity, endogenous dynamics (Keynes, Soros, etc.) and the excess volatility puzzle quite far from the EMT picture Understanding the determinants of impact is crucial: From a theoretical point of view: why do price changes? Do they reflect some underlying «fundamental» price? From a practical point of view: price impact can be a large fraction of transaction costs 3

4 2. MARKET MICROSTRUCTURE Microstructure: Bid/Ask quotes (limit orders) and trades (Market orders) Ask-Bid=Spread S Market makers post quotes They face (possible) adverse selection In the «old days» ( ): S= 70 bp = 0.7% In present electronic markets: S= a few bps (competition!) 4

5 2. MARKET MICROSTRUCTURE Markets are hide and seek games and liquidity is inherently fragile (cf below) «Liquid markets» are actually not liquid: Total turnover in a day: 0.5% of Mcap But total volume in OB: only 10-5 of Mcap Hide and Seek game Adverse selection: Hint to why is liquidity so low 5

6 2. MARKET MAKING IS HARD Market makers attempt to earn the spread but lose adverse impact Simple maths: sell at 101, buy at 99: good but sell at 101 and buy back at 102 if the price went up in the meantime? Cost of a market order: Average impact (response) 6

7 2. MARKET MAKING IS HARD Gain of a (simple) market making strategy With inventory control 7

8 2. MARKET MAKING IS HARD Gains of a (simple) market making strategy with inventory control Note 1: In «equilibrium»: S and R are proportional Note 2: HF mechanically increases MM profits Note 3: HF vs low latency 8

9 2. MARKET MAKING IS HARD Gains of a (simple) market making strategy with inventory control The «blue» line: an equilibrium market ecology 9

10 2. MARKET MAKING IS HARD Spread and Response (impact) are indeed proportional And vol (per trade) and Response as well: Vol is mostly due to trades, little by news jumps with no trades So vol and spread are prop. too: spread (or taxes) influence vol! 10

11 11 2. MARKET MAKING IS HARD So: Market makers attempt to earn the spread but lose adverse impact : profits are necessarily small in a competitive liquidity providing setting Market makers try to anticipate the flow and avoid adverse selection any blip in perceived risk scares them away and leads to increased spreads (and this may cascade!) Spread is the friend, volatility is the foe: MM strategies tend to reduce volatility (this is in fact their very raison d etre!), but unexpected volatility makes the job hard liquidity is SMALL! All of the above was true in the old days, still true with HFT, except that liquidity providing in the old days was much better

12 12 2. HFT: THE NEW MARKET MAKING Market making is essential to markets how much is it worth? Market makers, much as HF machines, never knew anything about they just try to avoid adverse selection The salesman knows nothing about what he is selling, save he is charging a great deal to much for it (Oscar Wilde) Market making is a crazy job, mostly mechanical much better done by machines. (Also true of order execution). HFT/Computers allow reduce adverse selection and thus spreads Note: Some «aggressive» strategies are classified as HFT but should rather be seen as «low latency» (turnover is ~2/day)

13 2. HFT: THE NEW MARKET MAKING Estimated profits of HFT: 0.5 bp /daily transactions = 2B$/year for all US stocks (ballpark, decreases with time) Is it expensive? Not any more (cf.! S=70 bp in the past) Is it stable? Maybe not, but was it ever before? Note: As a general rule, more efficiency often leads to more fragility (small profits=less risk taking=more overreaction) 13

14 HFT & MARKET STABILITY: SOME EMPIRICAL FACTS Stocks daily turnover has only increased by a factor ~2 since 95 Spreads have noticeably decreased (70 bp 2 bp) Jump frequency has not increased (but scales as ) - has not increased (cf Filimonov-Sornette) Hawkes process description: : Branching ratio 14

15 3. MARKET IMPACT AND THE TRUE COST OF TRADING How much does one pay for transacting? Naïve answer from above: a fraction of the bid-ask spread True for small trades but as soon as size Q is substantial, impact costs become dominant (and >> than naively thought!) [BARRA 97(!), Almgren, Engle, JPM, DB, LH, CFM ] true liquidity and very little by microstructure (at variance with single orders) Order of magnitudes: for Q=2% of daily volume: Cost = 1 bp * 2% * sqrt(2%) = bp Impact is unavoidable and much larger than spreads! 15

16 IMPACT OF METAORDERS Empirical result: a square-root (non additive) impact! Independent of: Markets, epoch, style of trading (LO/MO), execution time, 16

17 IMPACT OF METAORDERS Why is impact a square-root? 17 Small liquidity Market fragility

18 IMPACT OF METAORDERS Why is «latent» volume linear in distance from price? A dynamical theory of liquidity : 18

19 IMPACT OF METAORDERS Why is «latent» volume linear in distance from price? A numerical «agent based» model of liquidity (Note 1: results robust viz changes of microstructural rules) : Note 2: slow decay of impact and «true» costs Cf: 19

20 CONCLUSION: INHERENT MARKET FRAGILITY