Title: Empirical Market Microstructure
1Empirical Market Microstructure
- Rotman School Distinguished Lecture Series
- March 2008
- Ingrid M. Werner
- Martin and Andrew Murrer Professor of Finance
- Fisher College of Business, The Ohio State
University
2Empirical tests of MM models
- Dealer markets
- Theory of Dealer Markets
- Cross-sectional spreads
- Inventory management
- Interdealer trading
- Asymmetric Information/Strategic Trading
- Transaction Costs
- Intraday patterns
- Spread decomposition
- Order processing
- Inventory
- Asymmetric information
- Price process
3The Inventory Control Hypothesis
- Objective
- To understand how market prices arise given the
nature of order flow and the market-clearing
protocol. - Focus
- The liquidity provider
- Specialist
- Dealer
- Market Maker (MM)
4Garman (1976)Bankruptcy Risk
- Can a market maker who posts bid and ask prices
and faces random asynchronous arrival of buy and
sell orders avoid bankruptcy? - That is, can he/she avoid running out of cash
and/or stock? - How should he/she set bid and ask prices to
maximize the chances of avoiding bankruptcy? - Gamblers ruin problem
- Key insightA positive spread is necessary to
avoid going bankrupt with probability one
5Garman (1976)Bankruptcy Risk
- Key assumptions
- Single risk-neutral monopolistic dealer
- sets the bid price, pb, and ask price, pa (only
once) - receives orders
- clears trades
- holds stock and cash
- Maximizes expected profits, tries to avoid
bankruptcy - Order-arrival processes
- one unit of stock
- Poisson arrivals
- Sells (?a(pa), ?a(pa) 0)
- Buys (?b(pb), ?b(pb) 0)
- Waiting time between order arrivals is
exponentially distributed - Buy and sell orders are driven by independent
processes - No herding or information driven trading
- Black-box, only depends on price, p
6Garman (1976)Bankruptcy Risk
- Key insights
- Order arrivals will affect both the inventory of
stock and cash - Let Ic(0) be the initial inventory of cash and
Is(0) be the inventory of stock. - Let R(t) (Q(t)) be the probability that the
dealer will run out of stock (cash).
- Note that pa gt pb is required for ?apa gt ?bpb
and ?alt ?b to hold - The dealer has to post a positive spread to avoid
failure with probability equals one. - No matter what prices the dealer sets, he will
fail with probability gt 0.
7Garman (1976)Bankruptcy Risk
- How are prices set?
- Raise ask and buy orders slow down
- Raise bid and sell orders arrive more rapidly
- To achieve zero-drift in inventories of cash and
stock, the dealer needs to set prices such that
arrival rates are equated - ?a(pe)?b(pe) ?e gt profits 0
- ?a ?b ? gt profits gt 0
- Shortcomings
- Dealer cannot change prices over time
- No link between prices and the inventory position
of the dealer
Dealer seeks to maximize shaded area
?b(pb)
pa
pe
pb
?a(pa)
?
Arrival rate
?e
8Amihud and Mendelson (1980) Inventory Management
- Generalizes Garman (1976) by modeling a profit
maximizing, risk neutral monopolistic dealer who
changes prices over time. - Punt on the bankruptcy problem by assuming that
inventory, I, is finite and bounded from below I
in (-K, L), L lt 8. - Translate the Poisson arrival process of Garman
(1976) into a birth and death process - This is a semi-Markov decision process where the
state variable is the inventory of stock, Ij, at
hand. - The decisions are to set bid, pb(Ij), and ask,
pa(Ij), prices. - Dealer controls the arrival rates directly as a
function of the state variable, Ij. - Key insightsA positive spread arises due to
dealer market power.Market prices will fluctuate
in part due to dealer inventory control.
9Amihud and Mendelson (1980)Inventory Management
- Optimal bid and ask prices are monotone
decreasing functions of the dealers inventory
position - The dealer has a preferred inventory position, -K
lt I lt L - The preferred inventory position is a function of
the variability of order arrival rates (cushion) - Positive spread arises because of market power
- The spread is increasing in the distance between
the dealers current inventory and his desired
inventory - The mid-quote is not always equal to the true
value of the stock! - The spread is wider than the spread that the
Garman specialist would set. - Competition would drive this spread to zero
- Empirically testable predictionsMarket prices
will fluctuate in part because of inventory
controlPrice effects of inventory imbalances are
transient
A2
A1
I1 gt I2
B2
B1
10Amihud and Mendelson (1980)Inventory Management
pa
?
pa
pb
?
pb
Inventorylevel
I
11Stoll (1978)Inventory Risk
- Dealer as a supplier of immediacy.
- Dealers are regular risk-averse traders.
- A dealer is a trader who voluntarily alters his
portfolio away from the optimal portfolio to
accommodate other traders demands. - Dealers require compensation for doing so.
- The compensation for risk is the bid-ask spread.
- Earlier work modeled risk-neutral traders (Amihud
and Mendelson (1980) and Garman (1976)) and the
spread was the result of monopoly power. - Key insightThe spread is compensation for the
risk the dealer takes on by tilting his portfolio
away from the optimal portfolio
12Stoll (1978)Inventory Risk
- Costs for providing immediacy
- Holding costs associated with suboptimal
portfolio. - Order-processing costs.
- Asymmetric information costs.
- Dealer is assumed to have exogenous beliefs about
the true price of the asset, and its true rate of
return. - Returns are normally distributed.
- Utility is negative exponential gt CARA
- Dealer maximized terminal wealth.
- Sets prices for one transaction (buy or sell).
- Inventory is financed by borrowing/lending at the
risk-free rate. - Market for supplying liquidity services is
competitive.
13Stoll (1978)Inventory Risk
- Let Qi be the true value of a trade in stock i
and Ci be the present dollar cost to the dealer
of trading Qi -
- Expanding both sides (Taylor series expansion),
dropping terms of order higher than two, setting
Rf 0, and simplifying the resulting expression
for the cost to the dealer of trading per unit of
Q is
14Stoll (1978)Inventory Risk
- Where z is the dealers coefficient of relative
risk aversion, Qp is the true dollar value of the
dealers optimal inventory, sip is the covariance
between the rate of return on stock i and the
optimal portfolio and si2 is the variance of
stock is return. - So, the cost function depends on
- Dealer wealth and risk preferences
- The level of the dealers optimal inventory
- The size of the trade in stock i
- The variance of the stock and its covariance with
the dealers optimal inventory.
15Stoll (1978)Inventory Risk
- If the dealer quotes symmetric quantities on the
bid and the offer, Qi, it is easy to show that
the spread relative to the true price will be -
- The spread is thus linear in trade size.
- The dealers inventory will affect where the bid
and ask prices are placed, but not the spread. - Empirically testable predictionSpread is
linearly increasing in trade size and in the
volatility of the stock
16Ho and Stoll (1981)Dynamic Model
- Extends the intuition of Stoll (1978) to a
multi-period framework in which both order flow
(Poisson) and returns are stochastic. - Dealers are risk-averse and maximize utility of
terminal wealth. - Model is solved using finite horizon dynamic
programming to characterize the dealers optimal
pricing policy. - Characteristics of the solution
- The spread includes an adjustment for risk that
depends on dealer risk aversion, the size of the
transaction, and the risk of the stock. - Transactions uncertainty per se does not affect
the spread. - In the limit there is only a risk-neutral
(monopoly) spread. - Dealers optimal pricing depends on the time
horizon (dealer becomes less risk averse over
time). - The spread is independent of the inventory level.
- Key insightThe spread arises due to market
power plus an adjustment for inventory risk plus
an adjustment for the risk from a suboptimal
asset allocation.
17Ho and Stoll (1983)Multiple Dealers
- How do dealers decide on optimal prices in a
setting with competition (e.g., Nasdaq, Forex,
OTC markets)? - Extension of Ho and Stoll (1981) to a setting
with multiple competitive dealers that can trade
with the public as well as with each other
(interdealer trading). - Public orders go to dealer with the best price,
and are randomly allocated if several dealers
quote the best price. - The order size is fixed.
- Dealer wealth and inventories are public
information. - Inventory is more risky than if the dealer had an
exclusive franchise. - Each dealers strategy depends on the strategies
of all other dealers. Need symmetry to
facilitate solution. - Key insight (English auction)Optimal strategy
is for the dealer who is the most eager to sell
(buy) to place his ask e above (bid e below) the
second most eager dealer to sell (buy).The
market spread depends on risk aversion, order
size, and riskiness of the stock and on the
distribution of inventories.
18Ho and Stoll (1983)Multiple Dealers
- As in previous models, the inventory will play a
key role. - Inventories dictate the reservation values for
each dealer, i.e., they determine the prices
(pa1ar, pb1-br) at which the dealer is
indifferent between trading and not trading. -
- Since inventories are public information,
everyone knows each others reservation values. - It is possible to order dealers in terms of their
eagerness to buy (lowest inventory) and to sell
(highest inventory).
19Ho and Stoll (1983)Multiple Dealers
- It is not necessarily optimal to quote
reservation values. - Suppose we order inventory levels (public
information) from the lowest to the highest I1
lt I2 lt lt IN-1 lt IN - It follows from our reservation quotesb1 lt b2 lt
lt bN-1 lt bNaN lt aN-1 lt lt a2 lt a1
Dealer N shades her ask by aN-1-aN -e
1a1
1a2
1aN-1-e
1aN-1
1aN
1-b1
1-b2
1-b2e
1-bN-1
1-bN
Dealer one shades his bid by b2-b1 -e
If you are the marginal dealer on either side,
set prices e away from the reservation quotes of
the second most desperate dealer..
20Ho and Stoll (1983)Multiple Dealers
- There is no guarantee that the optimal policy we
just derived will result in a market ask price
that exceeds the bid! - It depends on the support of the inventory
distribution, and on the other parameters (A, s2,
Q) - Introduce a round of interdealer trading prior to
customer order arrival
1a1
1a2
Dealer N shades her ask by aN-1-aN -e
1-b2e
1-b1
1-b2
1aN-1
1aN
1aN-1-e
Dealer one shades his bid by b2-b1 -e
1-bN-1
1-bN
Absent interdealer trading, optimal spread would
be negative, leavingroom for arbitrage by
customers...
21Ho and Stoll (1983)Multiple Dealers
- Implications (Empirically testable)
- As long as there are more than 2 dealers,
interdealer trading may occur. - Dealers will use interdealer trades to reduce
some of the discrepancies in inventories by
trading with each other. - They will then compete with each other for the
arriving customer orders (second price auction) - Quotes will be less spread out in the public
market than if we do not allow interdealer
trading - Market spread is declining in the number of
dealers - Problems
- The assumption of full transparency of
inventories is unrealistic! - Dealers also need to know A, subjective s2, etc
- Model is static and order flow is naïve
22Biais (1993)Incomplete Transparency/Fragmentation
- In practice, we cannot easily track dealer trades
or when a particular dealer quotes get hit
(fragmentation). - Can the model of Ho and Stoll (1983) be relaxed
to deal with incomplete transparency? - How would dealers optimally set their quotes if
they do not know each others inventories? - Biais (1993) assumes that dealers do not know
each others inventories, but they do know the
distribution from which each dealers inventory
is drawn. - Quotes are sealed-bid Dutch auction
- Key insightsDealers will act strategically and
shade their prices.The spread is wider than
under full transparency.Competition will reduce
the room for strategic behavior gt spreads
narrow when there are more dealers.
23Biais (1993)Incomplete Transparency/Fragmentation
- Game
- Traders decide whether or not to become a dealer
- Dealers receive inventories I in -R, R, iid
- Market order arrives, B/S with equal probability,
- Best quotes gets the order
- Final value of the asset is realized
- Solution is by backwards induction
- Lucky for us, it starts with Ho and Stoll (1981,
1983) solution for reservation quotes. - Difference is that the dealer does not know the
reservation values of his/her competitors (since
inventories are not public information) - Dealer needs to figure out the probability that
his ask (bid) price is lower (higher) than that
of all other dealers (could get the trade) - Bayes-Nash equilibrium concept
24Biais (1993)Incomplete Transparency/Fragmentation
- Assume that the inventory distribution is
uniform. - Optimal bid (1-bi) and ask (1ai) quotes are
Dealers shade their quotes Shading is increasing
in A, s2, R Shading is decreasing in the
numberof dealers Dealer spreads are decreasing
in the number of dealers
Dealer will post an ask price as if his
inventory is Ii lt Ii
1/2R
-R
R
Ii
0
Ii
(Ii-(-R)) divided by N determines shading
25Empirical Predictions Inventories
- Predictions from theory
- Arrival rate of orders is price elastic
- Quotes should reflect inventory position of
dealers. - The rate at which dealer trade inventory
imbalances depends on - risk aversion
- the number of dealers
- risk of the stock
- time
- There is a target inventory level and spreads
widen as we move away from this target inventory
level. - The market spread should be increasing in the
size of the customer trade - The market spread should be increasing in the
volatility of the stock - The market spread should be decreasing in the
number of dealers
26Stoll (1978)Cost of Trading on Nasdaq
Order ProcessingCosts
Risk-aversion
- Sample of 2,508 stocks for 6 trading days
- Median dealers is 5
- Median spread is 3-4
- The number of dealers is increasing in liquidity
- Not all registered dealers are actively trading
- Empirical model for spreads (log-linear)
- Spreads are declining in dealers and liquidity!
- Spreads are increasing in proxy for adverse
selection!
spread
Turnover
Risk
Dealers
Wealth
Volume
Avg. daily inventory change
Concentration ratio
CAPM
MM Residual
27Stoll (1976)Nasdaq Dealer Inventory
- Start with our model of dealer spreads (Ho and
Stoll (1978)) - Add order processing and information costs (ad
hoq) - Assume a lot of symmetry and perfect competition
- Let ?Q be the change in aggregate inventory
(proxy order imbalance) - Let P be the midquote and r the midquote return
- Use data on 5 days for 2,052 stocks
lt0 Suggests passive inventory adjustment rgt0 gt
?Q lt 0 stabilizing
lt0 Suggests mean Reversion gt 8-10 days to
reverse Q gt Inventory risk
Dealers get taken by informed traders
28Hasbrouck and Sofianos (1993)Trades of NYSE
Specialists
- Inventory is sometimes negative (short positions)
- There is no obvious drift or divergence
- Patterns suggests rather tight inventory
management - The mean inventory is near zero
- Overnight positions are small Dealers tend to go
home flat - Liquid stock
- Large market capitalization
29Hasbrouck and Sofianos (1993)Trades of NYSE
Specialists
Discrete jump Offsetting position in other
stock?Options/derivatives? Overseas
traded? Illiquid stock Small capitalization
Wandering inventory Offsetting position in other
stock?Options/derivatives? Overseas
traded? Illiquid stockSmall capitalization
30Hasbrouck and Sofianos (1993)Trades of NYSE
Specialist
- Stationary? They have to be (at least mean
reverting) - Noisy data
- Misinterpreted
- Horizon not long enough
- How fast do specialists get rid of inventory
positions? Slow
31Madhavan and Sofianos (1993)Trades of NYSE
Specialists
- Why is the mean reversion so slow?
- Measurement issues?
- Study days with extreme price changes
- Evidence shows that specialists reverts large
inventory shocks quickly - 24/40 cases one day
- 36/49 cases by fourth day
- Hedging
- Evidence suggests not commonly used
- Correlated securities
- Correlation of inventories across stocks is on
average positive for 41 out of the 50 specialists
units - Speculation
- Would mess up our estimates?
32Hasbrouck and Sofianos (1993)Trades of NYSE
Specialists
- Are NYSE specialists profitable?
- On average, profits are negative
- Short and medium term profits are positive
- Long term profits are negative and very noisy
- Specialists make profits from the spread, but
there is also some evidence of profitable
speculation (informed trading) - Prices, trades, and inventories
- Most of the quote dynamics are attributable to
trades, with inventories contributing little
explanatory power. - Positions are not managed by adjustment of
publicly quoted bids and offers.
- But, how is then inventory management done?
- Selective nonpublic quoting
- Interdealer trading
- Market specific rules that allow dealers to
participate in trades without publicly signaling
through their quotes (NYSE).
33Madhavan and Smidt (1993)Inventories and Quotes
of NYSE Specialists
- Develop a structural model to estimate
- Mean reversion of inventories, ß
- ß -1 implies inventories reversed in one day
- Desired inventory level, Id
- Results
- ß is negative and significant for 8 out of 16
stocks - ß is on average -0.05 which implies a half life
of 49 days (even slower than HS (1993)) - Desired inventory is positive and significant for
12 out of 16 stocks! - Bad news
- Structural model did not manage to eliminate the
slow mean reversion - Allow for shifts in desired inventory
- ß is negative and significant for 13 out of 16
stocks - ß is on average -0.134 which implies a half life
of 7.3
34Madhavan and Smidt (1993)Quotes and Inventories
of NYSE Specialists
- Half lives vary significantly across stocks.
- Midquote-updates are modeled as a function of
- Shocks to order imbalances ()
- Inventory adjustment (-)
- If there is a positive shock to (buy) order
imbalances, midquotes increases gt Information? - If inventory levels increase, midquotes decrease
- Thus, MS (1993) find a strong link between
quotes and inventories
35Madhavan and Sofianos (1998)NYSE Specialist
Trades
- How does dealer activity vary across stocks?
- Liquidity
- Off-exchange competition
- Tick size (tick/P high gt more specialist
trading?) - Specialist participation rate (SP) is defined as
specialist purchases and sales, divided by total
purchases and sales. - Level SP(Liquidity (-), Off-exchange competition
(-), Tick/P (-), ) - What affects specialist trading over time in an
individual stock? - Inventory
- Spread
- Trade size
- Momentum
- Signed SP(Inventory(-), Spread (-), Trade Size
(), Momentum (-), )
36Madhavan and Sofianos (1998)NYSE Specialist
Trades
37Madhavan and Sofianos (1998)NYSE Specialist
Trades
38Hansch, Naik, and Viswanathan (1998)Dealer
Inventories in London
- Study normalized inventories
- Over 50 of large trades are taken by dealer with
extreme inventory - Dealers posting at the BBO attract more than
their expected share of order flow - There is a strong (?) link between quotes and
inventories
39Hansch, Naik, and Viswanathan (1998)Dealer
Inventories in London
- Model mean reversion as a piece-wise linear
regression - There is considerable variation in the speed of
mean reversion depending on how extreme the
position is in the first place. - Mean reversion is still relatively slow!
- One sigma shock
- Half-life 5 days (All)
- Half-life 3.5 days (Non-ADR)
- Five sigma shock
- Half-life 1.3 days (All)
- Half-life 1.2 days (Non-ADR)
- Problems
- Variable transformation
- Pre-arranged trades
40Reiss and Werner (1998)London Interdealer Trading
41Reiss and Werner (1998)London Interdealer Trading
42Reiss and Werner (1998)London Interdealer Trading
- Does risk-sharing motivate interdealer trading
(Ho and Stoll (1983))? - Methodology
- Inventory cycles
- Prepositioned/Anticipated trades
- ID trading represents roughly 25 percent of
volume - Active inventory management compared to HNV
(1998) - One (FTSE) to two days (non-FTSE)
- ID trading conforms to Ho and Stolls (1983)
hypotheses - ID are used to reduce inventory imbalances
- Roughly 80 of ID trades are in the unwinding
direction - Roughly 65 of ID trades have both dealers
unwinding - Dealers in ID trades have extreme inventories
- Position taking is profitable, and profits are
larger if customer trades are used to unwind the
inventory
43Additional Research on Dealer Inventories
- Naik and Yadav (2003) study London dealers
trading multiple stocks and find that dealers do
not seem to take a portfolio approach to their
risk exposure. - Lyons (1995) study one FOREX dealer and finds
strong evidence of inventory control. - Manaster and Mann (1996) study futures markets
and find patterns consistent with inventory
control. - Chakravarty and Li (2003) study futures markets
and find rapid mean reversion in the personal
inventory of dual traders. - Kavajez and Odders-White (2001) study NYSE
specialists and find no evidence that specialists
revise their price schedules in response to
changes in their inventory.
44Inventory Empirics Conclusion
- Spreads do seem to be consistent with models of
dealer markets in the cross-section - Mean reversion of inventories seems surprisingly
slow in many markets - Link between quotes and inventories is
surprisingly weak - Why is dealer inventory management so difficult
to measure? - Dealers have complex objectives
- Off-exchange trading
- Order preferencing, payment for order flow, soft
dollars, etc
45Dealer Market References
- Theory
- OHara, M., 1995, Market Microstructure Theory,
Chapter 2. - Hasbrouck, J., 2007, Empirical Market
Microstructure, Chapter 11. - Amihud, Y., and H. Mendelson, 1980, Dealership
markets Market making with inventory, Journal of
Financial Economics 8, 31-53. - Biais, B., 1993, Price formation and equilibrium
liquidity in fragmented and centralized markets,
Journal of Finance 48, 157-185. - Garman, M., 1976, Market microstructure, Journal
of Financial Economics 3, 257-275. - Ho, T., and H. Stoll, 1981, Optimal dealer
pricing under transactions and return
uncertainty, Journal of Financial Economics 9.
47-73. - Ho, T., and H. Stoll, 1983, The dynamics of
dealer markets under competition, Journal of
Finance 38, 1053-1074. - OHara, M., and G. Oldfield, 1980, The
microeconomics of market making, Journal of
Financial and Quantitative Analysis 21, 361-376. - Stoll, H., 1978, The supply of dealer services in
securities markets, Journal of Finance 33,
1133-1151.
46Dealer Market References
- Empirical
- Bacidore, J. and G. Sofianos, 2002, Liquidity
provision and specialist trading in NYSE-listed
non-U.S. stocks, Journal of Financial Economics
63, 133-158. - Hansch, O., Naik, N., and S. Viswanathan, 1998,
Do inventories matter in dealership markets?
Evidence from the London Stock Exchange, Journal
of Finance 53, 1623-1656. - Hasbrouck, J., and G. Sofianos, 1993, The trades
of market makers An empirical analysis of NYSE
specialists, Journal of Finance 48, 1565-1593. - Madhavan, A., and S. Smidt, 1993, An intraday
analysis of daily changes in specialist
inventories and quotations, Journal of Finance
48, 1595-1628. - Manaster, S., and S. Mann, 1996, Life in the
pits competitive market making and inventory
control, Review of Financial Studies 6, 953-975. - Lyons, R., 1993, Test of microstructure
hypotheses in the foreign exchange market,
Journal of Financial Economics 39, 321-351. - Reiss, P., and I.M. Werner, 1998, Does risk
sharing motivate interdealer trading? Journal of
Finance 53, 1657-1703. - Stoll, H., 1976, Dealer inventory behavior An
empirical investigation of Nasdaq stocks, Journal
of Financial and Quantitative Analysis 11,
359-380.