Market Microstructure and High-Frequency Trading

Topic 3: Order Flow and Price Dynamics

Part I

Bo Liu, Ph.D

Huron at Western

Jan 23, 2025

[email protected]

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 1 / 85
What did we do last topic?

1 Discuss liquidity
2 Talk about different measures of:
spreads (quoted, effective, realized)
price impact
other measures
3 Estimators for missing data
Lee-Ready algorithm for estimating trade direction
Roll’s spread estimator requiring only price data

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 2 / 85
This topic

1 Beginning of a long discussion about determinants of the spread

2 Go in-depth with the relation between information and prices
3 What is informational efficiency?
4 Discuss Glosten and Milgrom’s model of information-based trading

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 3 / 85
 Information and Prices

This topic:

1 Information and Prices

2 Glosten-Milgrom model

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 4 / 85
 Information and Prices

Determinants of liquidity

Last topic: illiquidity ⇒ deviation of realized prices from the asset’s

fundamental value.
Source of illiquidity: market thinness (too few traders present in the
market at a given time).
If that was the only reason, trivial solution for efficiency: add many
dealers.
Unsurprisingly, this does not close the spread.
So what drives the spread?
To answer that, first talk about what drives the prices.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 5 / 85
 Information and Prices

Determinants of liquidity

Last topic: illiquidity ⇒ deviation of realized prices from the asset’s

fundamental value.
Source of illiquidity: market thinness (too few traders present in the
market at a given time).
If that was the only reason, trivial solution for efficiency: add many
dealers.
Unsurprisingly, this does not close the spread.
So what drives the spread?
To answer that, first talk about what drives the prices.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 5 / 85
 Information and Prices

Determinants of liquidity

Last topic: illiquidity ⇒ deviation of realized prices from the asset’s

fundamental value.
Source of illiquidity: market thinness (too few traders present in the
market at a given time).
If that was the only reason, trivial solution for efficiency: add many
dealers.
Unsurprisingly, this does not close the spread.
So what drives the spread?
To answer that, first talk about what drives the prices.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 5 / 85
 Information and Prices

Determinants of liquidity

Last topic: illiquidity ⇒ deviation of realized prices from the asset’s

fundamental value.
Source of illiquidity: market thinness (too few traders present in the
market at a given time).
If that was the only reason, trivial solution for efficiency: add many
dealers.
Unsurprisingly, this does not close the spread.
So what drives the spread?
To answer that, first talk about what drives the prices.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 5 / 85
 Information and Prices

Determinants of liquidity

Last topic: illiquidity ⇒ deviation of realized prices from the asset’s

fundamental value.
Source of illiquidity: market thinness (too few traders present in the
market at a given time).
If that was the only reason, trivial solution for efficiency: add many
dealers.
Unsurprisingly, this does not close the spread.
So what drives the spread?
To answer that, first talk about what drives the prices.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 5 / 85
 Information and Prices

Determinants of liquidity

Last topic: illiquidity ⇒ deviation of realized prices from the asset’s

fundamental value.
Source of illiquidity: market thinness (too few traders present in the
market at a given time).
If that was the only reason, trivial solution for efficiency: add many
dealers.
Unsurprisingly, this does not close the spread.
So what drives the spread?
To answer that, first talk about what drives the prices.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 5 / 85
 Information and Prices

Price determinants

Stock prices move around all the time

Media reports are full of ex post rationalizations
But these can be pretty arbitrary
What actually determines stock prices and their movement?

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 6 / 85
 Information and Prices

Price determinants

In principle: everybody agrees on assets’ future cash flows (≈value),

conditional on fundamentals. Broadly speaking, three reasons for trading:
Risk: Getting the right risk profile (relevant to trader)
(Funding) Liquidity: Trader needs liquid funds or has excess funds to
invest (relevant to trader)
Speculation: Different expectations of fundamentals because
different information (relevant to everyone)

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 7 / 85
 Information and Prices

Types of information

Different types of information (it’s a spectrum really)

Public information: asset price may move without trade
Private information possessed by some traders → reveal this
information through their trading
Draw a wedge between legal and academic (← our) definitions of
private information
Legal: insider info is that which is only available to a restricted number
of people. Trading based on insider info is illegal in most jurisdictions.
Academic: some agents may be better at analyzing publicly available
info, thus have better info about fundamentals.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 8 / 85
 Information and Prices

Types of information

Different types of information (it’s a spectrum really)

Public information: asset price may move without trade
Private information possessed by some traders → reveal this
information through their trading
Draw a wedge between legal and academic (← our) definitions of
private information
Legal: insider info is that which is only available to a restricted number
of people. Trading based on insider info is illegal in most jurisdictions.
Academic: some agents may be better at analyzing publicly available
info, thus have better info about fundamentals.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 8 / 85
 Information and Prices

Information and prices

“Fundamentally, in a system in which the knowledge of the

relevant facts is dispersed among many people, prices can act to
coordinate the separate actions of different people.” – Von Hayek

Fama’s efficient markets hypothesis: prices must be efficient.

I.e., they must reflect all available information.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 9 / 85
 Information and Prices

Information and prices

“Fundamentally, in a system in which the knowledge of the

relevant facts is dispersed among many people, prices can act to
coordinate the separate actions of different people.” – Von Hayek

Fama’s efficient markets hypothesis: prices must be efficient.

I.e., they must reflect all available information.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 9 / 85
 Information and Prices

Types of informational efficiency

There are different kinds of price efficiency:

Weak form: Prices reflect historic (price) information
Semi-strong form: Prices reflect all public information
Strong form: Prices reflect all public and private information
Strong is the kind of efficiency we had in mind in the previous lectures. It
arises in classical models in Economics (see General Equilibrium theory)

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 10 / 85
 Information and Prices

Issues with EMH

Issues with EMH:

No-trade theorem: If traders have private information and care only
about fundamentals, nobody should ever trade (Milgrom and Stokey
[1982])
GS informational paradox: (Grossman and Stiglitz [1980])
Suppose markets were informationally efficient, then prices would
reflect traders’ private information
Thus: no incentive to acquire information (if this is costly)

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 11 / 85
 Information and Prices

Issues with EMH

Issues with EMH:

Volatility: prices move a lot, such volatility cannot be justified by
public news alone
Process: how does information get incorporated into prices?
Suppose every agent has some private signal about the fundamental,
but also knows that the price incorporates all private info of all agents.
Then it’s optimal to ignore own signal, since it adds nothing to the
price signal. But if everyone ignores all signals, price cannot be
informative.
So we will not take EMH for granted, but rather see whether it arises in
models endogenously.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 12 / 85
 Information and Prices

Asset value

Let’s try to figure out how price efficiency looks like in math.
Information: Ωt captures the market’s (public) knowledge at time t –
ever more is known: Ωt+1 = (Ωt , It+1 ).
Market valuation (̸= price) of an asset = expectation of underlying
fundamental value v given information Ωt :

µt = E [v |Ωt ] .
P∞
Think of v as the sum of discounted cash flows: v = s=t δ s−t cs
(uncertain at t).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 13 / 85
 Information and Prices

Informational efficiency

Informational efficiency then corresponds to: pt = µt = E[v |Ωt ]

(Ωt is “market’s knowledge”, so efficiency understood as semi-strong)
Valuation only changes if new information arrives: ‘innovation in
value’ is a random variable: ϵt+1 = µt+1 − µt . Then

E[ϵt+1 |Ωt ] = E[µt+1 − µt |Ωt ]

= E[µt+1 |Ωt ] − E[µt |Ωt ]
= E[E[v |Ωt+1 ]|Ωt ] − µt
= E[v |Ωt ] − µt
= µt − µ t
= 0.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 14 / 85
 Information and Prices

Informational efficiency

Also, E[ϵs ϵt ] = 0, ∀s ̸= t.
Price innovation is equal to the valuation innovation:

pt+1 − pt = µt+1 − µt = ϵt+1

Thus
E[pt+1 |Ωt ] = pt
When we have informational efficiency, the price is a martingale

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 15 / 85
 Information and Prices

How is information incorporated into prices?

Prices are set by market makers (liquidity suppliers = sell side):

dealers
other speculators (anyone placing limit orders)

They gather information Ωt from:

public announcements, news, etc.
observing the order flow (insofar as some orders are driven by private
information)

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 16 / 85
 Information and Prices

Recall Conditions for EMH to Hold

Risk neutral and competitive market making:

zero expected profits
All information is public:
investors are not more informed than market makers
No order processing costs for market makers:
no need to cover such costs in price
Then: “zero expected profits” implies

bid price = ask price = µt

⇒ zero bid-ask spread

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 17 / 85
 Glosten-Milgrom model

This topic:

1 Information and Prices

2 Glosten-Milgrom model

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 18 / 85
 Glosten-Milgrom model

Glosten and Milgrom [1985]

All models are wrong; some models are useful.

– George Box

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 19 / 85
 Glosten-Milgrom model

GM85: Overview

The simplest model explaining spread through private information.

Spread is driven by adverse selection.
Dynamic model, periods t = 1, 2, . . . :
Though we will be analyzing the stage game for a given period —
essentially static.
Two players in every period:
Trader and Dealer.
Dealer: long-lived; Trader: new every period.
Trader can be informed or not.
One asset with fundamental value v (unknown):
Common belief: v ∼ F (v ).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 20 / 85
 Glosten-Milgrom model

GM85: Model

Trader: is either a speculator or a noise trader, and can submit a market

order dt ∈ {1, −1} to buy or sell one unit of the asset with fundamental
value v (or do nothing, dt = 0).
Speculator (probability π): has private information about v .
We usually assume the speculator simply knows v (not much changes if
he only has a noisy private signal about it).
Risk-neutral, chooses market order dt to maximize expected profit:

dt · (v − pt )

Noise trader (probability 1 − π): no private information about v ;

trades for other reasons (hedging, liquidity).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 21 / 85
 Glosten-Milgrom model

Dealer (Market Maker)

Risk neutral
Willing to trade exactly one unit (buy/sell/no trade) each period.
Sets bid and ask prices (for a single unit).
Quotes price before seeing trade (limit order).
Does not know whether the trader is a speculator or noise trader (but
knows π).
Expected profit from trade is:

E[−dt (v − pt )].

Competitive: prices = expected asset value conditional on

information.
Trading is sequential: market orders are served one by one.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 22 / 85
 Glosten-Milgrom model

Aside on Dealers

“For each security in which a member is registered as a Market Maker,

the member shall be willing to buy and sell such security for its own
account on a continuous basis during regular market hours and shall
enter and maintain a two-sided trading interest (“Two-Sided Obligation”)
that is identified to the Exchange as the interest meeting the obligation
and is displayed in the Exchange’s quotation montage at all times.”
– Nasdaq Rule 4613

“Minimum requirements: at least 85% of the time, at most 4% bid-

offer spread, order size at least worth 4,000 euros”
– Nasdaq OMX Helsinki

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 23 / 85
 Glosten-Milgrom model

GM85: Model

Asset value:
Random asset value v drawn from distribution
Assume v is the fundamental (terminal) value
Speculators know v perfectly (not much changes if they don’t)
Equilibrium:
An equilibrium consists of bid and ask prices and speculator’s strategy
They must be such that: (i) prices are competitive (zero profit for
MM), (ii) speculator best-responds to prices (maximizes expected
gain).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 24 / 85
 Glosten-Milgrom model

Market making

Dealer quotes bid and ask prices on one unit

Can revise prices between each incoming trade
Quoted ask price at only relevant if next incoming trader decides to
buy
Dealer’s payoff in this case is given by
E[at − v |Ωt−1 , Buy ] = at − E[v |Ωt−1 , Buy ]
Same for bid bt ; payoff: E[v |Ωt−1 , Sell] − bt
(Note payoffs above rely on risk-neutrality)
Perfect competition among dealers implies zero expected profit ⇒ ask
price and bid price are

at = E[v |Ωt−1 , Buy ];

bt = E[v |Ωt−1 , Sell].

Notice that both sides of the equality depend on prices

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 25 / 85
 Glosten-Milgrom model

Informative Order Flow: Glosten-Milgrom Model

Simplest example: high-low fundamental value and unit trade size

Security’s value: (
1
vH with prob. 2,
v= 1
vL with prob. 2.

Order flow at time t: dt = 1 (buy order) or dt = −1 (sell order).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 26 / 85
 Glosten-Milgrom model

Informative Order Flow: Glosten-Milgrom Model

Market maker’s estimate of the security’s value v as of time t:

pt = µt ≡ E(v |Ωt )

Market maker’s expected value after observing t th order:

µt ≡ E(v |Ωt−1 , dt )
|{z}
"belief" at t

= θt v H + (1 − θt )v L
| {z } | {z }
probability of high value at t probability of low value at t

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 27 / 85
 Glosten-Milgrom model

GM85: Model

At each time t, a single trader arrives on the market:

With probability π: the trader is informed and knows v .
Buys 1 unit if the value is higher than the ask (v = vH > a).
Sells 1 unit if the value is lower than the bid (v = vL < b).
With probability 1 − π: the trader is uninformed ("noise" or
"liquidity" trader) and does not know the true value of the security.
1
Buys or sells 1 unit with probability 2
each.

The market maker ignores whether the trader is informed or not: this
creates asymmetric information ⇒ adverse selection.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 28 / 85
 Glosten-Milgrom model

Learning from Order Flow

The distribution of the order flow depends on the true value of the
asset:
1+π 1−π
Pr(dt = 1|vH ) = and Pr(dt = −1|vH ) =
2 2
1−π 1+π
Pr(dt = 1|vL ) = and Pr(dt = −1|vL ) =
2 2
Hence, dealers can learn about the value of the asset from the order
flow.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 29 / 85
 Glosten-Milgrom model

Bid and Ask Prices

Competition among risk-neutral market makers pushes their expected
profits down to zero.
⇒ Ask and bid prices equal their estimate of the security’s value
conditional on a buy or a sell order :

at = µ+
t ≡ E(v |Ωt−1 , dt > 0)

bt = µ −
t ≡ E(v |Ωt−1 , dt < 0)

Intuitively:
E(v |Ωt−1 , dt > 0) > E(v |Ωt−1 , dt < 0)
Key point: dt can be used to forecast v because the distribution of dt
depends on v .
The bid-ask spread the consequence of buy and sell orders convey
different messages.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 30 / 85
 Glosten-Milgrom model

Determinants of the Bid-Ask Spread

Consider ask firstly, ask Price at Time t:

In a perfectly competitive market with risk-neutral dealers:
Dealers set the ask price at to ensure zero expected profits.
Losses on trades with informed investors are offset by profits from
liquidity traders.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 31 / 85
 Glosten-Milgrom model

Determinants of the Bid-Ask Spread

Market Maker’s Considerations:

1 Informed Investor:

Sells to an informed investor when v = v H .

Loss: at − v H .
Probability: πθt−1 , where:
π: Probability of informed trading.
θt−1 : Probability the asset has value v H .
2 Liquidity Trader:
Sells to a liquidity trader, with no new information since t − 1.
Loss: at − µt−1 (market maker’s estimate of the final value).
Probability: (1 − π)/2, where:
1 − π: Probability of liquidity trading.
1/2: Equal likelihood of buy/sell by the liquidity trader.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 32 / 85
 Glosten-Milgrom model

Transaction Probabilities and Underlying Values

Figure: Transaction Probabilities and Underlying Values

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 33 / 85
 Glosten-Milgrom model

Dealer’s Expected Net Profit at Ask Price (at )

Using the probabilities, the dealer’s expected net profit from transactions at
the ask price at is given by:

θ π · (a − v H )
|t−1 {zt }
Expected profit from trading with informed customer

+ 0 · (at − v L )
| {z }
No profit from trading with v L (informed customer)
1
+ (1 − π) · (at − µt−1 )
2
| {z }
Expected profit from trading with uninformed customer
 
1
+ (1 − θt−1 )π + (1 − π) · 0
2
| {z }
Probability of no ask-side customer (no trade)

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 34 / 85
 Glosten-Milgrom model

Setting the First Ask Price in the Trading Day

We focus first on the special case of a market maker who is choosing, at

time 0, the first ask and bid quotes of the trading day:
Ask is set conditional on the arrival of a buy order (q1 > 0).
Dealers are competitive: expected profits per share are zero:
π 1−π
(a1 − v H ) + (a1 − µ0 ) = 0
2 2
π
⇒ a1 = µ0 + π
2
1−π
(v H − µ0 )
2 + 2

= µ0 + π(v H − µ0 )

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 35 / 85
 Glosten-Milgrom model

First Ask and Bid Prices in the Trading Day

Dealers’ initial estimate of v is its unconditional mean:

vH + vL
µ0 =
2
So:
π H
a1 = µ0 + π(v H − µ0 ) = µ0 + (v − v L )
2
Symmetrically on the bid side:
π H
b1 = µ0 + π(v L − µ0 ) = µ0 − (v − v L )
2

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 36 / 85
 Glosten-Milgrom model

First Bid-Ask Spread in the Trading Day

Bid-ask spread at t = 1:

S1 = a1 − b1 = s1a + s1b
π H π
= (v − v L ) + (v H − v L )
2 2
= π(v H − v L )
The spread just covers dealers’ losses with informed traders.
Cost of adverse selection is ultimately borne by liquidity traders:
Markets are often organized/regulated to alleviate informational
asymmetries.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 37 / 85
 Glosten-Milgrom model

Ask Price at Any Time t in the Trading Day

Ask is set conditional on arrival of a buy order (qt > 0).

So that dealers’ expected profits per share are zero:
1−π
πθt−1 (at − v H ) + (at − µt−1 ) = 0
2
πθt−1
⇒ at = µt−1 + (v H − µt−1 )
πθt−1 + 1−π2

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 38 / 85
 Glosten-Milgrom model

Ask Price, Bid Price, and Spread at Any Time t

But since:
µt−1 = θt−1 v H + (1 − θt−1 )v L
We have:
πθt−1 (1 − θt−1 )
at = µt−1 + (v H − v L )
πθt−1 + (1 − π)/2

Symmetrically:

πθt−1 (1 − θt−1 )
bt = µt−1 − (v H − v L )
π(1 − θt−1 ) + (1 − π)/2

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 39 / 85
 Glosten-Milgrom model

Ask Price, Bid Price, and Spread at Any Time t

Spread:
St = 2πθt−1 (1 − θt−1 )×
 
1 1
+ (v H − v L )
2πθt−1 + 1 − π 2π(1 − θt−1 ) + 1 − π

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 40 / 85
 Glosten-Milgrom model

Determinants of Bid-Ask Spread: Insights

Key Factors Influencing the Spread:

Proportion of informed traders (π): Higher π leads to larger spreads.

Volatility of the security’s value (v H − v L ).

Dealers’ beliefs (θt−1 ) about the value of the security:
Maximum spread occurs when θt−1 = 0.5, reflecting high uncertainty.
When θt−1 = 0.5, dealers face maximum uncertainty, making them
highly sensitive to new orders and leading to significant revisions.
Spread decreases as θt−1 approaches 1 or 0, indicating confidence in
v H or v L .
When θt−1 is close to 1, dealers are confident in the asset’s value,
resulting in minimal updates to their estimates (at ≈ µt−1 ).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 41 / 85
 Glosten-Milgrom model

Likelihood of informed trading as: π

Figure: Likelihood of informed trading

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 42 / 85
 Glosten-Milgrom model

Uncertainty on the final value (maximal for θ = 0.5)

Figure: Uncertainty on the final value

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 43 / 85
 Glosten-Milgrom model

How Dealers Revise Their Quotes

Dealers’ quotes (at , bt ) depend on their beliefs about the security’s

value.
Beliefs are updated based on order flow:
θt+ : Probability of v = v H after a buy order.
θt− : Probability of v = v H after a sell order.
Dealers use Bayes’ Rule to revise these probabilities dynamically.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 44 / 85
 Glosten-Milgrom model

Updating Probabilities

After a buy order:

θt+ = Pr(v = v H |Ωt−1 , dt = +1)

Using Bayes’ Rule:

Pr(dt = +1|v = v H ) · Pr(v = v H )

θt+ =
Pr(dt = +1)

Substituting:
(1 + π) 21 · θt−1
θt+ =
πθt−1 + (1 − π) 12

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 45 / 85
 Glosten-Milgrom model

Updating Probabilities

After a sell order:

θt− = Pr(v = v H |Ωt−1 , dt = −1)

Using Bayes’ Rule:

(1 − π) 21 · θt−1
θt− =
π(1 − θt−1 ) + (1 − π) 12

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 46 / 85
 Glosten-Milgrom model

Revisions in Quotes

Updated dealer expectation upon receiving a buy order at time t:

µ+ + H + L
t = θt v + (1 − θt )v

Revision of value after a buy order:

πθt−1 (1 − θt−1 ) H
µ+
t − µt−1 = 1
(v − v L ) = sta
πθt−1 + (1 − π) 2

Revision of value after a sell order:

−πθt−1 (1 − θt−1 )
µ−
t − µt−1 = (v H − v L ) = stb
π(1 − θt−1 ) + (1 − π) 12

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 47 / 85
 Glosten-Milgrom model

Implications of Quote Revisions

Information from order flow:
Buy orders indicate a higher likelihood of v = v H .
Sell orders indicate a higher likelihood of v = v L .
Bid-ask spreads:
St = sta + stb
πθt−1 (1 − θt−1 ) H
sta = (v − v L ),
πθt−1 + (1 − π) 12
πθt−1 (1 − θt−1 )
stb = (v H − v L )
π(1 − θt−1 ) + (1 − π) 12
sta and sta are changes to dealers’ estimates of asset value, proves that
order flow is a source of information.
Bid-ask spreads reflect compensation for adverse selection risks.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 48 / 85
 Glosten-Milgrom model

Dealers’ Quotes and Their Roles

The deviation of the ask (bid) quote from the previous estimate of the
stock value µt−1 plays two roles:
Allow the dealer to earn a revenue to cover the expected losses made
by trading with informed investors:

at − ut−1 > 0, bt − ut−1 < 0 (bid-ask spread)

Update the previous quotes in line with the informational content of a

buy (sell) order:

ut+ − ut−1 > 0, ut− − ut−1 < 0 (quote updating)

The two coincide, since:

at = ut+ , bt = ut−

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 49 / 85
 Glosten-Milgrom model

Key Insights and Applications

Dealers’ quote adjustments provide insight into information

asymmetry in markets.
Larger spreads arise when uncertainty (θt−1 ≈ 0.5) is high.
Practical implications:
Useful for understanding market reactions to order flow.
Relevant for regulatory policies addressing adverse selection costs.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 50 / 85
 Glosten-Milgrom model

Price Discovery: Transaction Price

Transaction prices are determined by the executed orders:

(
at = µ +
t if dt = +1,
pt = −
bt = µt if dt = −1.

Hence,
pt = µt = θt v H + (1 − θt )v L ,
where (
θt+ if dt = +1,
θt =
θt− if dt = −1.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 51 / 85
 Glosten-Milgrom model

Semi-Strong EMH

Prices reflect all available information to market makers:

Information at t = Ωt−1 (past orders) + dt (current order).

Semi-Strong EMH:
Transactions occur at fair values given public information.
The order flow, Ωt = {d1 , d2 , . . . , dt }, is observed.
In this model, the semi-strong EMH holds true.
Key Insight:
Bid-ask spread reflects the incorporation of order flow into price
discovery.
Temporary deviations are a result of information asymmetry, not
inefficiency.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 52 / 85
 Glosten-Milgrom model

Strong Version of EMH

The strong form of the Efficient Market Hypothesis (EMH) states:

Transactions always occur at fair values given all public and private
information.
For instance:

v = v H =⇒ all trades occur at prices v H .

Challenge:
Seems impossible for markets to be efficient in this sense.
How can prices reflect information unavailable to all participants?

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 53 / 85
 Glosten-Milgrom model

Price Dynamics with Full Information (π = 1)

Suppose all traders are informed (π = 1), quotes at t = 1:

a1 = v H , b1 = v L .

When v = v H , informed traders buy:

First transaction occurs at p1 = a1 = v H .
Dealers infer θ1 = 1 (high value) and adjust quotes for subsequent
trades.
Implication:
Dealers learn from private information contained in the order flow.
The process is called price discovery.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 54 / 85
 Glosten-Milgrom model

Price Discovery and Dynamics

Dealers’ beliefs about the asset’s fair value evolve as they process
order flow:
θt → 1 as informed traders dominate.
Speed of price discovery depends on the proportion of informed traders
(π):
Higher π: Faster convergence to true value v H or v L .
Lower π: Slower adjustment.
Price discovery can happen if π > 0.
Without informed trading:
Order flow is balanced (50% buys, 50% sells).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 55 / 85
 Glosten-Milgrom model

Simulating Price Discovery

Example setup:
v H = 102, v L = 98, µ0 = 100, θ0 = 0.5, π = 0.3.
True value is v H .
First two orders buy, the third is sell:

θ1 = 0.65, θ2 = 0.77, θ3 = 0.65...

p1 = 100.6, p2 = 101.10, p3 = 100.6...

Simulation:
Generate 10 random sequences of 100 orders (dt = {+1, −1}) with
same π = 0.3).
Update beliefs (θt ) and transaction prices (pt ).
Measure price discovery:

PDt = (pt − v H )2

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 56 / 85
 Glosten-Milgrom model

Example: simulating the dynamics of dealer beliefs

Figure: Dearlers beliefs

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 57 / 85
 Glosten-Milgrom model

Simulation: dynamics of the (squared) pricing errors

Figure: PDt = (pt − v )2 , here v = v H

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 58 / 85
 Glosten-Milgrom model

Simulation: price discovery for different values of π

Figure: Squared pricing error for π = 10%, 50% and 90%

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 59 / 85
 Glosten-Milgrom model

Illiquidity and Informational Efficiency

Hence, when π increases, illiquidity (bid-ask spread) increases, but the

speed of price discovery increases as well:
1 There is a trade-off between liquidity and informational efficiency, at
least in the short run.
2 More frequent informed trading widens bid-ask spreads and tends to
make the market illiquid, but it also speeds price discovery.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 60 / 85
 Glosten-Milgrom model

Quotes and Transaction Prices

Figure: Quotes and Transaction prices for AGF

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 61 / 85
 Glosten-Milgrom model

Price Movements and Order Flow

The model provides a framework for explaining the effect of order flow on
price movements. Recall that:

µ+ a
t = µt−1 + st , µ− b
t = µt−1 − st

These equations can be written more compactly as:

µt = µt−1 + s(dt )dt

where: (
s a , if dt = +1
s(dt ) ≡ tb
st , if dt = −1

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 62 / 85
 Glosten-Milgrom model

Order Flow and Informative Market Orders

The dealers’ estimate of the value of the security after the t-th transaction,
µt , depends on the direction of the order flow. Market orders are
informative: (
s a , if dt = +1
s(dt ) ≡ tb
st , if dt = −1
Dealers set their quotes at time t to bracket their prior estimate of the
security’s value, µt−1 :
at = µ+t = µt−1 + st
a

bt = µ− b
t = µt−1 + st

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 63 / 85
 Glosten-Milgrom model

Numerical Illustration
Consider the following assumptions:
1 1
v H = 102, v L = 98, θ0 = , π= .
2 2
On these assumptions, we find:
a1 = 101, b1 = 99.
3
If the first order is a buy (d1 = +1), the probability θ1 is updated to 4
and µ1 = 101.
Dealers then revise their quotes to:
a2 = 101.6, b2 = 100.
Conversely, if the first order is a sell (d1 = −1), the probability θ1 is
updated to 14 and µ1 = 99.
Dealers revise their quotes to:
a2 = 100, b2 = 98.4.
Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 64 / 85
 Glosten-Milgrom model

Dynamics

First period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 65 / 85
 Glosten-Milgrom model

Dynamics

Second period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 66 / 85
 Glosten-Milgrom model

Dynamics

Third period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 67 / 85
 Glosten-Milgrom model

Dynamics

Fourth period: sell

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 68 / 85
 Glosten-Milgrom model

Dynamics

Fifth period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 69 / 85
 Glosten-Milgrom model

Dynamics

Sixth period: sell

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 70 / 85
 Glosten-Milgrom model

Dynamics

Seventh period: sell

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 71 / 85
 Glosten-Milgrom model

Dynamics

Eighth period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 72 / 85
 Glosten-Milgrom model

Dynamics

Ninth period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 73 / 85
 Glosten-Milgrom model

Dynamics

Tenth period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 74 / 85
 Glosten-Milgrom model

Dynamics

Eleventh period: sell

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 75 / 85
 Glosten-Milgrom model

Dynamics

Twelfth period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 76 / 85
 Glosten-Milgrom model

Dynamics

Thirteen period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 77 / 85
 Glosten-Milgrom model

Dynamics

Fourteen period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 78 / 85
 Glosten-Milgrom model

Dynamics

Fifteen period: buy

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 79 / 85
 Glosten-Milgrom model

Dynamics Evolution and Trade Prices

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 80 / 85
 Glosten-Milgrom model

Dynamics Evolution of Market Maker’s Belief

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 81 / 85
 Glosten-Milgrom model

Correlation Between Transaction Prices and Order Flow

All transactions take place either at the ask price or the bid price, so:

µ+ a
t = µt−1 + st , µ− b
t = µt−1 + st

µt = µt−1 + s(dt )dt

(
s a , if dt = +1
s(dt ) ≡ tb
st , if dt = −1
pt − pt−1 = µt − µt−1 = s(dt )dt .

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 82 / 85
 Glosten-Milgrom model

Correlation Between Transaction Prices and Order Flow

The difference between the t − 1 th transaction and the t th

transaction is determined entirely by the direction of the t th order
(dt ).
The price innovation ϵt is equal to s(dt )dt , which explicitly relates the
change in price to the order flow.
Variance of price changes: var(∆pt ) = var(s(dt )dt ).

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 83 / 85
 Glosten-Milgrom model

Return Volatility and Bid-Ask Spread

Return volatility at time t is determined by:

The size of the bid-ask spread s(dt ).
The uncertainty of the direction of order flow dt .
As quotes become more accurate, the spread narrows, and return
volatility decreases.
Observations in equity markets:
Volatility and bid-ask spreads decline mid-session but increase near the
market close.
Private information influences order flow and may cause temporary
widening of spreads.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 84 / 85
References I

L. R. Glosten and P. R. Milgrom. Bid, ask and transaction prices in a

specialist market with heterogeneously informed traders. Journal of
Financial Economics, 14(1):71–100, Mar. 1985. ISSN 0304-405X. URL
https://doi.org/10.1016/0304-405X(85)90044-3.
S. J. Grossman and J. E. Stiglitz. On the impossibility of informationally
efficient markets. The American economic review, 70(3):393–408, 1980.
Publisher: JSTOR.
P. Milgrom and N. Stokey. Information, trade and common knowledge.
Journal of Economic Theory, 26(1):17–27, Feb. 1982. ISSN 0022-0531.
doi: 10.1016/0022-0531(82)90046-1. URL https://www.
sciencedirect.com/science/article/pii/0022053182900461.

Bo Liu, Ph.D Market Microstructure and High-Frequency Trading Jan 23, 2025 85 / 85