Transcript Should we trust the dismal scientists in white coats?

Experimental Markets
Chris Starmer
TSU Short Course in Experimental and Behavioural
Economics, 5-9 November 2012
Background
• In 2002, Vernon L. Smith Shared
Nobel prize in Economic Science
• For work developing experimental
approaches to study of markets
– Discovery of remarkable properties
• This (more-or-less) started of one
of the major research programmes
in experimental econ.
• I talk about this today
Visit - www.nobelprize.org/nobel_prizes/economics/laureates/
Route Map
The approach Smith pioneered uses an
experimental technique called
Induced Value Methodology
A technique for setting up experimental markets
Today I will
Introduce the method
Show some classic – often replicated - results
(performance of competitive equilibrium)
Extension to asset markets
What is an Experimental Market?
• Experiment in which:
– Some experimental subjects assigned roles of buyers
and/or sellers
– There are ‘goods’ that can be traded
• Often ‘induced value’ tokens (See later)
• Markets (usually) set up so that:
– there are potential gains from trade
• Experimenter controls rules of trade
• Observes what happens
V.Smith 1989, J.EconPerspectives
• Market experiment is conjunction of:
– Environment
• Participants, endowments, preferences…
– Institution
• Rules of trade, who can do what, when …..
– Behaviour
• Experiments often involve
– Creation/manipulation of Environ/Institution
– With view to observing consequent behaviour
A Classic Market Experiment
Vernon Smith (1962) J. Pol. Econ.
“An Experimental Study of
Competitive Market Behaviour”
Environment (Smith 1962)
• Subjects divided at random into (Buyers, Sellers)
• Each buyer Bi given a ‘value’:
– MAX PRICE (vi) they can pay for unit of (fictitious)
commodity
• Each seller Sj given a ‘cost’:
– MIN PRICE (cj) at which they can sell a unit of commodity
• vi and cj - private information for individuals
• If trade takes place between i and j at price p:
Bi = vi – p
sj = p – cj
• Smith asked participants to try to max profit
– Note: hypothetical in this early experiment
Interpretation of Payoffs
• In the aggregate:
– the values given to buyers define a demand
function
– so, for any potential price, the values determine
the maximum quantity that can could be
purchased
Illustration........
Suppose there were
6 buyers with
max prices:
v1 = 1
v2 = 2
v3 = 3
v4 = 4
v5 = 5
v6 = 6
Price
Buyer ‘Values’ and Demand
Price 7
Price 6
Price 5
Price 4
Price 3
Price
Price
2
1
1
2
3
4
5
6 Demand
Seller ‘costs’ and market supply
Price
Suppose there were
6 sellers with
Price 7
min prices:
c1 = 2
c2 =3
.
.
.
c6=7
Price
Price
Price
Price
Price
6
5
4
3
2
1
1
2
3
4
5
6 Supply
Buyer values and
seller costs used in
this way INDUCE
demand and supply
schedules in an
experimental market
Price
Supply and Demand
7
6
Supply
5
4
3
2
1
Demand
1
2
3
4
5
6
Price
Equilibrium
Applying standard
competitive
7
equilibrium
6
theory, we can
5
identify the
Equil.
4
Price
equilibirum
3
(range) for Price
2
and Quantity
Supply
Demand
1
1
2
3
4
5
6
Back to Smith 1962…..
Supply and Demand conditions
in Smith 1962
• Smith reports results of running several
market sessions
• Demand and Supply conditions vary
• Here are D and S schedules from one
market
Demand and Supply induced in ‘Test 1’
(Source: Smith, 1962, p 113)
Question
Should we expect equilibrium in this
experimental market?
Not Necessarily
• “The mere fact that […] supply and demand schedules
exist in the background of a market does not guarantee that
any meaningful relationship exists between those
schedules and what is observed in the market they are
presumed to represent. All the supply and demand
schedules can do is set broad limits on the behaviour of the
market.
• (Vernon Smith, 1962, p.115)
• In other words, D and S determine the set of
feasible trades
• What will arise as the actual pattern of trading is
an open question
Institutions Matter
• Simple thought experiment demonstrates that
institutions matter….
• Consider an extreme case where the rules of the
experiment are
– Everyone sits in isolation
– No possibility to communicate with other potential
traders
• In this case, the institutional arrangements
inhibit/prevent trade:
– No prices formed
– Zero quantity traded
Smith’s 1962: uses a ‘Double Auction’
• Sequence of trading periods (5–10 mins)
• Experimenter opens/closes trading periods
• At any time during trading period B (or S) are free to
make verbal offers (hence DA)
– These must respect max values (min costs)
• Improvement rule for new offers
– Higher bid/lower offer than current best
• Any S ( or B) can accept an existing offer to buy (or
sell) s.t. cost (value) constraints
• Acceptances result in binding contracts
– B/S pair drop out of market once party to a contract
• Trading period continues until no further contracts
are being made
Realisticness Smith’s markets
see Smith 1962 pp.115- 116
Features like world
• Traders ignorant of
each others values
• Learn about other’s
values by observing
others’ willingness to
trade
• Small numbers of
traders
Features unlike world
S and D held constant over
trading periods in
experiment
Smith’s (1962) Results: “Test 1”
(Source: Smith, 1962, p 113)
Deviation from eq. Price
In early periods
‘Convergence’ toward
Equilibrium in later periods
Variation in S and D Schedules
1. Changing slopes of Demand and
Supply functions
Smith 1962: Changing slopes of D and S
Double-Auction Markets: Experimental Results
(Source: Smith, 1962)
Rapid convergence for (relatively) flat supply and demand
functions
Double-Auction Markets: Experimental Results
(Source: Smith, 1962)
Convergence is more erratic for steeper schedules
(note (relatively) higher α values for this market)
Impact of Surplus Distribution
Producer’s Surplus > Consumer’s Surplus
(Source: Smith, 1962)
Notice that there is convergence from below CE
This pattern is typical of markets where producers’ surplus is greater
The pattern tends to be reversed when the dist. of surplus is reversed
Variation in S and D Schedules
Demand Expansion
Demand Expansion
Baseline demand condition
for 4 periods
In period 4, buyers get
new private max prices
inducing demand expansion
Note different equilibrium
predictions
(Source: Smith, 1962)
Results of demand shift – upwards sloping supply
(Source: Smith, 1962)
Rapid convergence – Periods 1 –4
Rapid (overshooting) reaction to shift in demand
‘Quick’ (2 period) adjustment to new equilibrium
Surprised by convergence in
Smith’s experiment?
Perhaps you should be?
• Consider some of key
assumptions of CE
models
–
–
–
–
Many Buyers
Many Sellers
Price Taking
Perfect Info
• Compare with Smith’s
markets
– Handful of traders
– Price Making
– Imperfect info
Each trader knows only their
own value. Individuals do not
know the general supply and
demand conditions
Commenting, years later on accumulated evidence from
experiments using DA markets, Smith argues….
• “There are no experimental results more important or more
significant than that the information specifications of
traditional competitive price theory are grossly overstated.
The experimental facts are that no double auction trader
needs to know anything about the valuation conditions of
other traders, or have any understanding or knowledge of
market supply and demand conditions, or have any trading
experience (although experience may speed convergence)
or satisfy the quaint and irrelevant requirement of being a
price ‘taker’ (every trader is a price maker in a double
auction).”
– Vernon Smith (Quote reproduced in Holt, p370)
Session 5 – Part II
Asset Market Experiments
Asset Markets
I discuss application of induced value to study
asset markets.
I discuss a landmark study Plott, C and S.
Sunder (1982)
– “Efficiency of Experimental Security Markets
with Insider Information…..”, J. Political Econ.
• And some subsequent literature
Characteristics of Asset Markets
• ‘Assets’
– Value may depend on uncertain state of nature
– May deliver multi-period returns (dividends)
• Experimental assets typically have one or both of these
features
• Traders
– buy and sell with view to profit
– potential for ‘speculation’
• Often asymmetric information
– ‘Insiders’ know more about value of asset (or state of
nature)
Two (caricatured) views on
efficiency of asset markets
Label these loosely as:
The efficient market hypothesis
The speculation hypothesis
Efficient Market Hypothesis
• Asset prices tend to reflect all relevant
information
– determined by ‘fundamentals’
• (correspond with discounted present values)
– respond rapidly (if not instantaneously) to new
information
• Various Rational Expectations models in
this spirit
Speculation Hypothesis
• Roughly:
– Pursuit of speculative profits can lead prices to
deviate from fundamentals
– May create potential for ‘bubbles’ and ‘crashes’
in stock prices
One rationale for asset market
experiments
• EMH is difficult to test with field data:
– Researchers don’t know discounted PVs
– Researchers don’t know what relevant info is
– Researchers can observe if responses to good and
bad news have right sign, but don’t know if price
levels reflect fundamentals
Advantage of Experiments
In principle, Experimenter can:
• control fundamentals
– e.g. implement asset with PV known to experimenter
• implement markets with Rational Expectations
Equilibria known to experimenter
• manipulate information structures
– E.g. create insider information and observe impact
Issues
• Two what extent, and under what conditions, do
markets:
– disseminate and aggregate dispersed information?
– achieve rational expectations equilibria?
• To what extent, and under what conditions, do we
see out of equilibrium phenomena?
– e.g. speculative bubbles and crashes
A Classic Asset Market
Experiment
Plott and Sunder 1982
Plott and Sunder - Basics
• Run 5 multi-period markets in which:
– At start of each period ‘investors’ endowed with assets
(2 units each)
– Holder at end of period is paid dividend
– Dividend value varies for three sets of investors
(I,II,III)
• Hence potential gains from trade
– For every investor, dividend D(S) also depends on state
of nature S in {X,Y}
– State of nature, varies randomly across market periods
(induced) payoff function
• Investors can hold, buy or sell assets
• Payoff (individual i) in each period (t) determined
as:
$it = Ci + Rit + nitD(S)it – Eit
$it= dollar earnings
Ci = cash endowment
Rit= revenue from sales
nit = holdings of assets (end of period t)
D(S)it= dividend
Eit= i’s expenditure on purchase of assets in period t
state of nature
• Determined randomly in each period (‘year’)
• All agents knew:
– State probabilities (via training
Insiders (and outsiders)
• In some periods, a subset of agents (usually
half of each group I, II, III) were informed
about state (‘insiders’)
– achieved by experimenter randomly selecting
state of nature from {X,Y}
– then distributing cards to all investors
• ‘Insiders’ received card specifying S (= X or Y)
• Uniformed investors received blank cards
P+S: Predictions
Consider two theoretical models
Private Information (PI)
Rational Expectations (RE)
Common assumptions:
Agents are risk neutral & follow simple decision rule:
– Buy assets if P < E(dividend)
– Sell assets if P > E(dividend)
PI: E(.) based on private information
Hence different E(.) for Insiders/Outsiders
RE: E(.) based on full information
Market behaves as if all agents know realised state
PI and RE
Lead to different predictions re:
–Market prices
–Pattern of asset holdings
Illustrate with P + S: “Market 3”
Dividends by trader group
Market 3
Prices will be bid up until
equal to highest dividend
for realised state
In RE model, all agents act as if
they know the realised true state
Hence, actual dividends determine
market outcomes
Final owners will be those
with the highest dividend
in the realised state
When state is Y:
Uniformed agents
Have highest (expected)
dividend
In PI model:
Informed agents act on actual
values for realised state
Uninformed act on Expected
Dividend in final column
When state is X:
Informed agents bid
price up to 400.
Only informed hold
Results for Market 3
No Insider Info
Results for Market 3
Note: because there are no informed traders,
so no real reason to expect RE to work
Natural benchmark is PI prediction: 220
Convergence to PI prediction
Periods 3 – 10
half traders are informed
When realised state is X: common predicted price PI and RE
PI/RE Price
realised state Y: different predicted prices for PI and RE
PI price
RE price
Markets
converge to
RE Price
What about the pattern of asset
holdings?
• Predictions (PI/RE) differ for state Y
• Experience Matters
– In early periods, allocation more consistent
with PI
– In later periods, allocations more consistent
with RE
• Uniformed agents (I,II) learn not to buy at prices
that seem attractive (given expected dividends)
P + S Conclusions
• “Given time and replication these markets behave
substantially as predicted by RE equilibrium
models. There seems to be no doubt that variables
endogenous to the operation of these markets
served to convey accurately the state of nature to
otherwise uninformed agents. We can conclude
that the RE models must be taken seriously as not
universally misleading about the nature of human
capabilities and markets” p692
Later Research
• Later research – explores factors
affecting efficiency of info
transmission
– Balance of informed vs uninformed
– Uncertainty re presence of informed
– When state of nature is revealed
– Experience of traders
– Number and Nature of assets traded
Hunting Bubbles
• Smith, Suchanek and Williams 1988
(“SSW”)
– “Bubbles crashes and endogenous expectations in
experimental spot asset markets”, Econometrica, 56,
1119–51.
Why Hunt Bubbles?
• SSW conjecture that real asset markets may
be prone to bubbles defined as:
– “trade in high volumes at prices that are
considerably at variance from intrinsic values”
p831.
• But - unclear how common bubbles are in
actual markets
– Because we don’t know intrinsic values
SSW Design
Two key distinguishing features compared to
earlier asset market studies.
– (eg. Plott and Sunder, 1982)
• Longer lived assets
– To allow ‘room’ for speculation
• Induce identical dividend structures across
market participants
- Only obvious motive for trade is speculative
Some generic design features
• Double auction markets
• Traders can switch between buy/sell
• 15 periods (mostly)
• At start each trader gets:
• Asset Endowment + Cash Endowment (CE)
• Assets give (risky) dividend each period
• Final payoff
• CE + cumul(trading profit (loss) + dividends)
Asset Value Structure
• at end of each period asset generates a
random dividend D
– D is random draw over (x1,x2,x3,x4)
– p(xi) = ¼ for all i
• if there are n periods:
– initial expected dividend E(D1) = nxi/4
– declines by xi/4 each period
• assets values same for all traders
Predictions
• Given the common value structure, known by all
traders
– if agents form expectations rationally
– and are risk neutral
• there are no gains from trade
• If there is trade
– a natural benchmark for the price would be the asset’s
‘fundamental’ value in each period T
• ie. E(DT)
Some stylised results
1. Typically high volume trading
2. Price bubbles and crashes are common
An illustrative market from SSW
Source - SSW p1130
An illustrative market from SSW
Source - SSW p1130
Conclusion
• Market experiments – one of the fist and
largest research programmes of exp. econ.
• Induced Value- Powerful tool for studying
behaviour of INSTITUTIONS
• Results show that sometimes, results of
econ theory work remarkable well
– Competitive equilibrium in double auctions
– In other places, disequilibrium phenomena are
common (Bubbles in asset markets)
• So, lot’s more to study here……
But for now, that’s all from me so
Didi madloba da
nakhvamdis!