chunyang tong sriram dasu information & operations management marshall school of business...
TRANSCRIPT
Chunyang TongSriram Dasu
Information & Operations Management
Marshall School of BusinessUniversity of Southern California
Los Angeles CA 90089
Dynamic Pricing under Strategic Dynamic Pricing under Strategic ConsumptionConsumption
A FrameworkA Framework
Consumers Strategic Consumers Strategic
The Seller Capacitated
No Yes
No
Yes X ( case IV)
Single priceis optimal (case II)
Price discriminatingis optimal (case I)
Literature (case III)
Strategic Consumers: Anticipate prices and buying strategies of other buyers
Are Buyer’s Strategic?Are Buyer’s Strategic? Empirical Evidence: “The price reduction Empirical Evidence: “The price reduction
occurrence can sometimes mean a more occurrence can sometimes mean a more reliable source to come back to, time and reliable source to come back to, time and time again.” (time again.” (http://www.buyersale.com/sale_info.htmlhttp://www.buyersale.com/sale_info.html ))
Experimental Evidence: Posted price Experimental Evidence: Posted price market buyers withhold demand (Ruffle, market buyers withhold demand (Ruffle, 2000)2000)
Problem SettingProblem Setting A Seller (risk-neutral monopolist) has A Seller (risk-neutral monopolist) has KK units of product units of product
to sell in finite horizonto sell in finite horizon A pool of A pool of NN risk-neutral consumers with heterogeneous risk-neutral consumers with heterogeneous
valuation. Commonly known is the cdf valuation. Commonly known is the cdf G(v)G(v), , Single-unit demand per consumerSingle-unit demand per consumer Consumers have anticipation of future prices and Consumers have anticipation of future prices and
maximize their expected surplusmaximize their expected surplus Excess demand is resolved via proportional rationing Excess demand is resolved via proportional rationing
(inefficient rationing) mechanism(inefficient rationing) mechanism Since initial capacity Since initial capacity KK is exogenously given, cost is is exogenously given, cost is
treated as sunktreated as sunk
Period 2 Period 1
Pricing Schemes & Information Pricing Schemes & Information StructuresStructures
Two pricing schemes: Two pricing schemes: Upfront pricing and contingent pricing Upfront pricing and contingent pricing
Two information structures: Two information structures: Common posteriors and common Common posteriors and common priorspriors
Common priors: buyer has a Common priors: buyer has a conditional probability distribution conditional probability distribution based on his own valuation and a based on his own valuation and a common priorcommon prior
Uncapacitated Seller, Uncapacitated Seller, Strategic ConsumersStrategic Consumers
Upfront Pricing Scheme ( P2, P1) Since consumers face no risk of stock-out, they simply choose
min(P2, P1)
Single-pricing Optimal
Randomized Pricing ( P2, F(P1)) Due to common knowledge of market information ( K,N,G(v)), Single Pricing Optimal
Upfront Pricing (Upfront Pricing (PP22,P,P11), ), deterministic demanddeterministic demand
With limited supply single pricing may not be optimal:
Example:
K=2, N=10, with valuation ( 100,40,35,30,28,26,25,23,21,20)
Optimal Single Price Scheme: P*=100, with revenue = 100
Two Price Scheme ( P2, P1)=(82, 20), with revenue = 82 + 20 = 102
Period 2 Period 1
Upfront Pricing Scheme, Upfront Pricing Scheme, Deterministic demandDeterministic demand
Two-price scheme is optimalTwo-price scheme is optimal
Price
time time time
Price Price
P3
P2
P1
Pc
Pc
Pc
P3
P2
P1
Lemma: The optimal pricing scheme consists of two prices ( P2, P1).The clearing price Pc is located in between.
Upfront Pricing Scheme, Upfront Pricing Scheme, Stochastic demand (common Stochastic demand (common
posteriors)posteriors)Consumers’ Symmetric Bayesian Nash Equilibrium Strategy:
Threshold Policy: Only buyers with valuation v y* will buy in the second last period. Others will defer to the last period.
y* solves the following equation: 2(y)(y – p2) = 1(y)(y- p1)
where:i(y) : probability of the buyer getting the object in period i.
Period 2 Period 1
A unique structure of equilibriumA unique structure of equilibrium
Conjecture: Threshold is uniqueConjecture: Threshold is unique Provide sufficient conditions for the Provide sufficient conditions for the
threshold to be uniquethreshold to be unique Numerically verified that it is unique Numerically verified that it is unique
for U(0,1)for U(0,1)
Upfront Pricing Scheme, Upfront Pricing Scheme, Stochastic demand (common Stochastic demand (common
posteriors)posteriors)
Computational result for Computational result for uniform distributionuniform distribution
For uniform distribution (0,1), N=5-50, K=1-(N-1), P2 is increasing Function of y.
Computational result for Computational result for uniform distributionuniform distribution
The more scarce the product is, the larger gap between prices
Impact on ProfitabilityImpact on Profitability
K/NK/N 1/41/4 2/42/4 3/43/4
Buyers’ Buyers’ strategy strategy consideredconsidered
Threshold value/PThreshold value/P22 0.7076/0.7076/
0.61290.61290.6235/0.6235/
0.46090.46090.5590/0.5590/
0.34600.3460
Total expected Total expected revenuerevenue
0.53200.5320 0.78690.7869 0.84740.8474
Ignoring Ignoring buyer’s buyer’s
strategystrategy
Actual y*/PActual y*/P22 0.9632/0.9632/
0.73790.73791/0.6821/0.68211
1/0.6551/0.65555
Total expected Total expected revenuerevenue
0.35860.3586 0.57250.5725 0.76800.7680
% of revenue % of revenue lossloss
32.6%32.6% 27.3%27.3% 9.4%9.4%( P1 is fixed at 0.3 )
Asymptotic ResultsAsymptotic Results Prices are monotone non-increasing Prices are monotone non-increasing Upfront pricing scheme leads to a Upfront pricing scheme leads to a
valuation-skimming process. It is valuation-skimming process. It is strategically equivalent to declining price strategically equivalent to declining price auction when the number of price changes auction when the number of price changes approaches infinity.approaches infinity.
When number of buyers approaches When number of buyers approaches infinity, a single price ( close enough to the infinity, a single price ( close enough to the upper bound of valuations) almost upper bound of valuations) almost guarantees a near-optimal profit.guarantees a near-optimal profit.
Contingent Pricing Scheme Contingent Pricing Scheme (common posteriors)(common posteriors)
Buyers and sellers have common Buyers and sellers have common knowledge knowledge
Seller determines price based on Seller determines price based on sales in previous periodsales in previous period
Contingent Pricing Scheme Contingent Pricing Scheme (common posteriors)(common posteriors)
Consumers in period 2 will buy immediately if and only Consumers in period 2 will buy immediately if and only if if
K
ii
i PxiPx1
*112
2 )()Pr()(
The curves of LHS and RHS have only one crossing point y*
The consumers’ equilibrium strategy is again a threshold policy
P2P1
RHS
LHS
y*
RHSofslopeijj
Ki
K
i
iN
kj
K
i
1
111
2 )Pr()Pr()Pr(
Impact on Profitability – Value of Impact on Profitability – Value of Dynamic PricingDynamic Pricing
Valuation uniform between 0 & 1K (number of units for sale) 5 4 3 2 1N (number of participants) 10 10 10 10 10
Single Period Optimal Revenue 2.26625 2.043808 1.719728 1.281258 0.715201Two Period Optimal Revenue 2.3330371 2.184288 1.860892 1.395448 0.779945% Change 2.95% 6.87% 8.21% 8.91% 9.05%
Single Period optimal Price 0.58 0.62 0.67 0.79 0.79
Two Period Pricing
Second Last Period Price 0.57 0.62 0.693 0.768 0.838Cut off (Buying Threshold) 0.8 0.74 0.79 0.83 0.88
Last Period Price When Inventory is1 0.576 0.548 0.6004 0.6391 0.69522 0.536 0.51 0.561 0.59763 0.504 0.481 0.52934 0.48 0.465 0.464
Contingent Pricing, Common Contingent Pricing, Common PosteriorPosterior
Prices may increasePrices may increase
Relative value of Dynamic Pricing Relative value of Dynamic Pricing depends on level of scarcitydepends on level of scarcity Best for “moderate” levels of scarcityBest for “moderate” levels of scarcity When N is very large in the limit a single When N is very large in the limit a single
price is adequate price is adequate
Extensions of Common Extensions of Common Posterior CasePosterior Case
Threshold policy and approach for Threshold policy and approach for computing optimal prices extend to:computing optimal prices extend to:
Multiple periodsMultiple periods New buyers entering each periodNew buyers entering each period Valuations changing over time Valuations changing over time
(provided expected valuations are (provided expected valuations are convex functions of time)convex functions of time)
Limitation of Common Limitation of Common Posterior ModelPosterior Model
Static pricing policy is near-optimal if:Static pricing policy is near-optimal if:1) the support of distribution is 1) the support of distribution is
bounded;bounded;2) Common posteriors;2) Common posteriors;3) Large number of buyers 3) Large number of buyers
To relax the assumption of common To relax the assumption of common posteriors, we can assume just common posteriors, we can assume just common priors on distribution of distributionspriors on distribution of distributions
Common PriorsCommon Priors
Distribution on distributionsDistribution on distributions (i) is the pdf for G(i) is the pdf for Gii(v) (distribution of (v) (distribution of
observed valuations)observed valuations)
Posterior distribution of each buyers Posterior distribution of each buyers depends on his/ her observed valuedepends on his/ her observed value
Assumption: GAssumption: Gii(v) G(v) Gjj(v), if i > j, where (v), if i > j, where i and j are the observed valuations of two i and j are the observed valuations of two buyers.buyers.
Example of Common PriorsExample of Common Priors
Prior distribution: N(Prior distribution: N(, , pp)) Buyer with observed valuation Buyer with observed valuation vv, believes , believes
that true mean that true mean ’ = ’ = vv + + , , where where is N(0 , is N(0 ,ee).). The posterior distributions are: N(The posterior distributions are: N(vv
where where vv = v*( = v*(22
pp/(/(22ee++22
pp)) + )) + *(*(22ee/(/(22
ee++22pp))))
= = 22ee22
pp/(/(22ee++22
pp)) ( The more you value the product, the ( The more you value the product, the
more you believe others value)more you believe others value)
Common PriorsCommon Priors
If , then a threshold If , then a threshold policy is a symmetric Nash Equilibrium.policy is a symmetric Nash Equilibrium.
),()( vi NvG
Work in ProgressWork in ProgressTerminal consumption: uncertain valuation until the final period
Capacity Control along with pricing Seller can strategically reduce supply
Multiple unit purchase Multi-firm competition
Experimental Studies