《Queueing Economic and Revenue Management 》课程中汇报的一篇论文,引用格式如下:
[1] Huang F , Guo P , Wang Y . Cyclic Pricing When Customers Queue with Rating Information[J]. Production and Operations Management, 2019, 28(10).
1. Introduction
cyclic pricing strategy 循环定价策略
两类顾客:
- sophisticated customer:知道 information,例如 service rate,service reward(精明客户)
- naive customer:只知道 average rating (幼稚客户)
provider 知道两类顾客的比例
M/M/1 queue
顾客的效用 consumptionutility=servicereward?waitingcostconsumption \ utility = service \ reward - waiting \ costconsumption utility=service reward?waiting cost
- waitingcostwaiting \ costwaiting cost 跟系统的拥堵程度(congestion level) 有关,拥堵程度又与 effective arrival rates by prices 有关。
根据历史顾客的效用会有一个评级(rating),average rating 会影响 incoming customer 的决策。
顾客的净效用 netutility=servicereward?waitingcost?pricenet \ utility = service \ reward - waiting \ cost - pricenet utility=service reward?waiting cost?price
- high-price 阶段:sophisticated customer 进,net utility = 0
- low-price 阶段:naive customer 进,net utility < 0
high-price 阶段 → less congestion → rating higher → average rating higher → lure naive customer → 价格虚高,naive customer 的效用<0,被overcharge
?never socially optimal
2. Literature Review
本文研究属于:1. customer’s strategic queueing behavior;2. provider’s information disclosure
Three streams of research:
- 关于information disclosure in queues,information包括:
- waiting time:本文中顾客获得的信息是异质的;队长是不可观测的。
- service reward:本文中有部分顾客是strategic的,另一部分是naive的
- service rate
- 关于 the bounded rationality of customers(顾客的有限理性)
- naive customer 是bounded-rational ,
- 同时本文还考虑了另一类sophisticated顾客
- 关于 pricing strategy for service facilities
3. Model Setup and Preliminaries
-
service times:iid,服从 rate为 μ\muμ 的指数分布(service rate = μ\muμ )
-
customer arrive(潜在顾客到达率):服从rate为 Λ\LambdaΛ 的泊松过程
-
service reward: RRR
-
price: ppp
-
waiting time(等待时间): W?exp?(μ?λ(p))W \sim \exp(\mu-\lambda(p))W?exp(μ?λ(p)) 服从指数分布,期望/平均等待时间 E(W)=1μ?λ(p)E(W)=\frac{1}{\mu - \lambda(p)}E(W)=μ?λ(p)1?
-
unit-time cost: ccc
-
waiting cost = (w+servicetime)?c(w + service \ time ) * c(w+service time)?c
-
effective arrival rate(join the queue 的有效顾客到达率): λ(p)\lambda(p)λ(p)
-
两类顾客比例:
- sophisticated:θ\thetaθ ,知道 μ\muμ 和 RRR => 潜在到达率 Λs=θΛ\Lambda_s = \theta \LambdaΛs?=θΛ
- naive: 1?θ1-\theta1?θ => 潜在到达率 Λu=(1?θ)Λ\Lambda_u = (1-\theta)\LambdaΛu?=(1?θ)Λ
把 TTT 时间划分成 NNN 个 phase:
(假设TTT 足够长,可以忽略phase 与phase之间的过渡阶段)
当N=1N=1N=1 时就退化为了 static 策略。
cyclic pricing strategy PTPTPT:
PT={(pi,Tpi)∣i=1,2,...,N}PT = \{ (p_i,T_{p_i})|i=1,2,...,N\} PT={
(pi?,Tpi??)∣i=1,2,...,N}
service provider 的长期目标:最大化长期平均利润(long-run average profit)=>其实也就是最大化利润率(profit rate):
max?PTΠ=1T[∑i=1Npiλ(pi)Tpi]\max_{PT}\Pi = \frac{1}{T}[\sum_{i=1}^{N}p_i \lambda(p_i) T_{p_i}] PTmax?Π=T1?[i=1∑N?pi?λ(pi?)Tpi??]
其中pip_ipi? 是阶段iii 的定价,λ(pi)\lambda(p_i)λ(pi?) 是阶段iii 的effective arrival rate,TpiT_{p_i}Tpi??是阶段iii 的时间长度。
上式等价于:
max?PTΠ=∑i=1Npiλ(pi)Lpi\max_{PT}\Pi = \sum_{i=1}^{N}p_i \lambda(p_i) L_{p_i} PTmax?Π=i=1∑N?pi?λ(pi?)Lpi??
其中 Lpi=TpiTL_{p_i}=\frac{T_{p_i}}{T}Lpi??=TTpi???
- 顾客的 consumption utility = R?cwR - cwR?cw (其中www 是waiting time)
- 平均评级 average rating —— η(PT)\eta(PT)η(PT) 会收敛到一个常数
- 每一阶段的 equilibrium arrival rate (平衡到达率):λ(pi)′s\lambda(p_i)'sλ(pi?)′s
两个步骤:1. 给定λ(pi)′s\lambda(p_i)'sλ(pi?)′s ==> 确定η(PT)\eta(PT)η(PT) ; 2. 给定 η(PT)\eta(PT)η(PT) ==> 确定 λ(pi)′s\lambda(p_i)'sλ(pi?)′s
step1:
step2:
-
naive customer 的 joining/balking decision(not have information about
service-related parameters RRR and μ\muμ , rely on the online rating information),要么全加入(=1),要么全不加入(=0)——信息缺失导致的羊群效应(herding behavior):
-
sophisticated customer, 是strategic 的,以一定的 joining probability 决定是否加入,目标是最大化期望效用 expected utility ,同时会考虑naive customer 的行为(即和δn(pi)\delta_n(p_i)δn?(pi?) 有关),令式(4)取得最大值的δs(pi)\delta_s(p_i)δs?(pi?) 即为sophisticated customer 的最佳决策:
显然,if the potential market size is large enough,如果潜在市场足够大,服务系统就一定会造成 congestion,sophisticated customer的最大效用就一定是0,所以求解上式就相当于只要求解:
可以解得:?λ(pi)=μ?cR?pi\lambda(p_i) = \mu - \frac{c}{R-p_i}λ(pi?)=μ?R?pi?c?,$\lambda(p_i) $ 见式(5).
下面为了方便起见,定义2个符号:
- VsV_sVs? 表示服务系统中只有 sophisticated customer 加入的情况下的 expected consumption utility;
- VnV_nVn? 表示服务系统中只有naive customer 加入的情况下的 expected consumption utility;
下面给出 命题1.
PROPOSITION 1. 当 service provider charges a price pi∈PTp_i \in PTpi?∈PT 的时候,the equilibrium arrival rates 满足下面的性质:
如果 pi>η(PT)p_i > \eta(PT)pi?>η(PT) ,定价大于average rating,所有的naive customer 都不加入(balk),即δn(pi)=0\delta_n(p_i)=0δn?(pi?)=0。系统中只有 sophisticated customers,这时候由?和式(5)可得,他们的join probability 为 δs(pi)=min?{1Λs(μ?cR?pi),1}\delta_s(p_i) = \min\{\frac{1}{\Lambda_s}(\mu - \frac{c}{R-p_i}), 1\}δs?(pi?)=min{ Λs?1?(μ?R?pi?c?),1}。effective arrival rate 为 λ(pi)=min?{μ?cR?pi,Λs}\lambda(p_i) = \min\{\mu - \frac{c}{R-p_i}, \Lambda_s \}λ(pi?)=min{ μ?R?pi?c?,Λs?}
如果 pi≤η(PT)p_i ≤ \eta(PT)pi?≤η(PT),定价小于average rating,所有的naive customer 都会加入(join),即δn(pi)=1\delta_n(p_i)=1δn?(pi?)=1。这时候sophisticated customers的joining probability由 average rating η(PT)\eta(PT)η(PT) 和 VnV_nVn? 决定:
- 若Vn≤pi≤η(PT)V_n ≤ p_i ≤ \eta(PT)Vn?≤pi?≤η(PT),则δs(pi)=0\delta_s(p_i)=0δs?(pi?)=0(由公式算出来是<0),所有的sophisticated customers都不加入(balk)。此时effective arrival rate 为 λ(pi)=Λn=(1?θ)Λ}\lambda(p_i) = \Lambda_n = (1-\theta)\Lambda \}λ(pi?)=Λn?=(1?θ)Λ};
- 其他情况由公式决定:
命题1 给出的结论是:
- naive customers 只有在定价不是很很高的情况下加入(pi≤η(PT)p_i ≤ \eta(PT)pi?≤η(PT));
- Sophisticated customers 只有在Vn≤pi≤η(PT)V_n ≤ p_i ≤ \eta(PT)Vn?≤pi?≤η(PT)的时候绝对不加入(joining probability 小于0,因为这个时候所有的naive customers 都加入了,此时的定价大于他们expected consumption utility VnV_nVn?,无法获得正的期望效益expected utility。
A Benchmark Case: θ=1\theta = 1θ=1 (全部都是sophisticated customer)
之前的文献有研究过,这里简单地列出在queue length is unobservable情况下的一些结论:
- M/M/1M/M/1M/M/1 queue,service provider 要决定price ppp 使他的profit-maximizing
- 所有customer都是strategic的,effective arrival rate λ\lambdaλ 要使得R?p?cμ?λ=0R-p-\frac{c}{\mu - \lambda} = 0R?p?μ?λc?=0
- maximizing provider’s profit 的问题可以转化为:
max?pλs.t.0≤λ≤ΛR?p?cμ?λ=0\max \ p\lambda \\ s.t. \ \ 0 \leq \lambda \leq \Lambda \\ R-p-\frac{c}{\mu - \lambda} = 0 max pλs.t. 0≤λ≤ΛR?p?μ?λc?=0
其最优解为:
即:所有customer都是strategic的情况下,
- 若 Λ≤λb\Lambda \leq \lambda_bΛ≤λb? ,则最优解是所有到达的潜在顾客都进入系统;
- 若Λ≥λb\Lambda \geq \lambda_bΛ≥λb? ,有部分顾客balk,the equilibrium effective arrival rate is λb\lambda_bλb?.
【注】:在这种Benchmark的设定下,profit-maximizing provider 的最优解 和 welfare-maximizing provider 的最优解相等。
4. Optimal Pricing Decision
4.1 Welfare Maximization
对于welfare-maximizing service provider来说,价格在customer and the provider之间的转化都是内部效应。Maximizing the long-run average social welfare 等价于 maximizing the expected total consumption utility:
为了方便起见,定义以下符号:
可以看到:
- θ<1/2\theta < 1/2θ<1/2 时:θ\thetaθ 增,Λ?\underline\LambdaΛ? 增,Λ?\overline \LambdaΛ 减, Λ^\hat \LambdaΛ^ 增;
- θ>1/2\theta > 1/2θ>1/2 时:θ\thetaθ 增,Λ?\underline\LambdaΛ? 减,Λ?\overline \LambdaΛ 增, Λ^\hat \LambdaΛ^ 减;
下面给出命题2:
PROPOSITION 2. The welfare-maximizing pricing strategy is static, that is, N = 1
(分情况讨论可以证明上述结论)
-
如果potential market size介于中间(Λ?≤Λ≤Λ?\underline \Lambda \leq \Lambda \leq \overline\LambdaΛ?≤Λ≤Λ) 并且 θ<1/2\theta < 1/2θ<1/2 时:
- Λ≤Λ^\Lambda \leq \hat\LambdaΛ≤Λ^ : all naive join(δn=1\delta_n=1δn?=1),all sophisticated balk(δs=0\delta_s=0δs?=0),如图
Case 1(a)
部分 - Λ>Λ^\Lambda > \hat\LambdaΛ>Λ^:all naive balk(δn=0\delta_n=0δn?=0),all sophisticated join(δs=1\delta_s=1δs?=1),如图
Case 1(b)
部分
- Λ≤Λ^\Lambda \leq \hat\LambdaΛ≤Λ^ : all naive join(δn=1\delta_n=1δn?=1),all sophisticated balk(δs=0\delta_s=0δs?=0),如图
-
除此之外,如果potential market size 很小(Λ≤Λ?\Lambda \leq \underline\LambdaΛ≤Λ?)或者很大(Λ≥Λ?\Lambda \geq \overline\LambdaΛ≥Λ), price 等于 all customers are sophisticated的情况(即benchmark case)
- Λ≤λb\Lambda \leq \lambda_bΛ≤λb?:all naive join(δn=1\delta_n=1δn?=1),all sophisticated join(δs=1\delta_s=1δs?=1),如图
Case 2(a)
部分 - λb<Λ≤Λ?\lambda_b < \Lambda \leq \underline\Lambdaλb?<Λ≤Λ? 或者 Λ?≤Λ≤Λ?\underline \Lambda \leq \Lambda \leq \overline\LambdaΛ?≤Λ≤Λ且1/2≤θ<11/2 \leq \theta < 11/2≤θ<1 时:all naive join(δn=1\delta_n=1δn?=1),sophisticated join with probability δs=λb?ΛnΛs\delta_s=\frac{\lambda_b - \Lambda_n}{\Lambda_s}δs?=Λs?λb??Λn??,如图
Case 2(b)
部分 - Λ≥Λ?\Lambda \geq \overline\LambdaΛ≥Λ: all naive balk(δn=0\delta_n=0δn?=0), sophisticated join with probability δs=λbΛs\delta_s=\frac{\lambda_b}{\Lambda_s}δs?=Λs?λb??,如图
Case 2(c)
部分
- Λ≤λb\Lambda \leq \lambda_bΛ≤λb?:all naive join(δn=1\delta_n=1δn?=1),all sophisticated join(δs=1\delta_s=1δs?=1),如图
可以看出,potential market size很小的时候(Λ≤λb\Lambda \leq \lambda_bΛ≤λb?),服务所有顾客;当potential market size介于中间(Λ?≤Λ≤Λ?\underline \Lambda \leq \Lambda \leq \overline\LambdaΛ?≤Λ≤Λ) 的时候,θ\thetaθ 对provider 的定价策略起了重要作用:
- 如果θ≥0.5\theta \ge 0.5θ≥0.5, the provider admits all naive customers into the system since the potential market size of naive customers is small(允许naive customer 全部进入,因为其市场总量就比较小);
- 如果θ<0.5\theta < 0.5θ<0.5,naive customer 市场总量比较大了,这时候 provider 要么只服务naive(Λ\LambdaΛ 较小,Λ≤Λ^\Lambda \leq \hat\LambdaΛ≤Λ^), 要么只服务sophisticated(Λ\LambdaΛ 较大,Λ^≤Λ\hat\Lambda \leq \LambdaΛ^≤Λ)。
Figure 2a 表明,social welfare 随着potential market size Λ\LambdaΛ 的增大而增大,会增大到其最大值 Λ=λb\Lambda = \lambda_bΛ=λb?(the socially desired arrival rate)并保持不变。但如果θ<0.5\theta < 0.5θ<0.5,Λ\LambdaΛ继续增大到Λ?≤Λ≤Λ?\underline \Lambda \leq \Lambda \leq \overline\LambdaΛ?≤Λ≤Λ 范围时,反而会 harms social welfare,因为这时候naive customer 出现羊群效应,导致 the effective arrival rate λb\lambda_bλb? 达不到最优值。
Figure 2b 表明,optimal price 随着potential market size Λ\LambdaΛ 的增大而减小,当 Λ\LambdaΛ 达到最优值 Λ=λb\Lambda = \lambda_bΛ=λb? 时降到最小值并保持不变。但如果θ<0.5\theta < 0.5θ<0.5,Λ\LambdaΛ继续增大到Λ?≤Λ≤Λ^\underline \Lambda \leq \Lambda \leq \hat\LambdaΛ?≤Λ≤Λ^ 范围时,optimal price 会开始降低,到Λ^\hat\LambdaΛ^ 点时突然跳到一个高点,之后在 Λ^≤Λ≤Λ?\hat \Lambda \leq \Lambda \leq \overline\LambdaΛ^≤Λ≤Λ 又开始降低,这是因为从serving naive customers only 阶段到 serving sophisticated customers only 阶段的转跳。
4.2. Profit Maximization
profit-maximizing provider 的优化问题和 前面的welfare-maximizing provider类似,唯一的不同点在于:把目标函数中的v(pi)v(p_i)v(pi?) 改为 pi′p_i'pi′? ,如下:
下面给出命题3 :
- potential market size存在一个threshold Λ~\tilde \LambdaΛ~,在这个threshold之上,provider 会选择cyclic pricing strategy;在这个threshold之下,provider的定价行为同
Proposition 2
中welfare maximization 的 static pricing strategy 一样。- The optimal cyclic pricing strategy is high-low cyclic:PT={(ph,L),(pl,1?L)}PT = \{(p_h, L),(p_l, 1-L)\}PT={ (ph?,L),(pl?,1?L)},其中php_hph?和plp_lpl?代表high and low price,LLL和1?L1-L1?L代表high price阶段的比例和low price 阶段的比例。Under the optimal cyclic pricing strategy, sophisticated customers join during the high-price phase and naive ones join during the low-price phase.
- The optimal low price satisfies pl?=η(PT)p_l^* = \eta(PT)pl??=η(PT), and the other two decision variables, (ph?,L?)(p_h^*, L^*)(ph??,L?), are given as follows.
- 【Ⅰ】. θ<1/2\theta < 1/2θ<1/2 且 $\tilde \Lambda < \Lambda < \ddot{\Lambda} $ 时:$\delta_s=\delta_n=1 $
- 【Ⅱ】. θ<1/2\theta < 1/2θ<1/2 且 $\Lambda \ge \ddot{\Lambda} $ 时,或者 θ≥1/2\theta \ge 1/2θ≥1/2 且 Λ>Λ~\Lambda > \tilde \LambdaΛ>Λ~时: $\delta_s=<1, \delta_n=1 $
从命题3中可以得出结论:
- 只有potential market size Λ\LambdaΛ达到一个特定值时,service provider 才会采取cyclic pricing strategy(Figure 3 右侧彩色部分),否则采取和welfare-maximizing provider中一样的static pricing strategy就可以了(Figure 3 左侧白色部分);
- 在采取cyclic pricing strategy的情况下:
- sophisticated customer 只在high-price 阶段进入,且expected net utility=0;
- naive customer 只在low-price 阶段进入,且expected consumption utility = VnV_nVn?。又因为最优的low price pl?=p_l^*=pl??=long-run average rating>Vn>V_n>Vn?,所以其expected net utility = Vn?pl?<0V_n - p_l^* < 0Vn??pl??<0总是负的。
- 所以即使naive customer 支付的是lower price,但还是不足以获取正的utility。
- 这种overcharging可能是因为the higher ratings generated from sophisticated customers, which boost up the average rating。
- 从Figure 3可以看出,Λ~\tilde \LambdaΛ~ 随着θ\thetaθ的增大而增大。即,系统中sophisticated customer 的比例越高,static pricing is more likely to be favored.
To thoroughly understand the mechanism underlying the profit gain,下面比较下 static and cyclic pricing strategies:
-
static pricing strategy: all customer surplus is internalized through pricing and goes towards the service provider’s profit(所有的消费者剩余都通过定价内部化,并转化为服务提供商的利润);因此,maximizing profit = maximizing welfare.
- 排队论中关于定价问题的经典文献已经表明,profit-maximizing service provider 的目标函数是关于 effective arrival rate 的凹函数,也就是说,如果能达到的话,λb\lambda_bλb? 就是provider的 best demand level(λb\lambda_bλb?是最佳的有效到达率)。
- 命题2 中已表明,在static pricing strategy下,当θ≥0.5\theta \geq 0.5θ≥0.5 (sophisticated customer的比例大于0.5)时,可以达到 最佳有效到达率λb\lambda_bλb? 。—— 可能你会认为在cyclic pricing strategy下的average arrival rate也应该等于λb\lambda_bλb? ,因为如果在high-price阶段小于$\lambda_b $ 的话,那在low-price 阶段应该可以大于λb\lambda_bλb? ,但不是这样的,看下面的推论。
-
cyclic pricing strategies:
- 推论Corollary 1表明,service provider 在both low- and high price phases都会故意(deliberately)只服务少于λb\lambda_bλb? 的顾客而获得更高的利润。
- 为什么不降低价格吸引更多的顾客呢?为什么原始的price–demand trade-off在这里不起作用呢?深入研究表明:
- low-price phrase,naive customer all join,并且获得负的 expected utility;这时候知道information 的sophisticated customer 永远不会join(all balk)。所以降价并不会增加需求,除非 naive customer 获得 non-negative expected utility。
- high-price phrase,降价可以增加需求,但是这样会造成系统 more congestion,从而会降低consumption utility 的评级,这样不利于在low-price phrase 对naive customer 的 overcharging strategy。所以,due to a higher profit margin from overcharging naive customers, the original price–demand trade-off is twisted(对naive customer 的overcharge 策略带来的高利润,导致价格-需求权衡发生了扭曲)。
- 综上,在两个阶段provider都选择只服务少于λb\lambda_bλb? 的客户。
Figure 4是代入具体的数值对cyclic pricing strategy的收益模式进行检验:
- (a)是最优利润,(b)是相比于static pricing 策略的利润增量,可以看到最大的利润增量平均在5.19%左右,最大可以达到11.22%。
Figure 5 是检验在cyclic pricing strategy中,how market size affects the optimal .
- (a)(b)都表明,在给定customer type composition parameter(θ\thetaθ) 的情况下,optimal prices 随着 potential market size Λ\LambdaΛ 的增大而减小。
- 对比虚线(optimal static price,即 the welfare-maximizing price)发现:
- when the market size is not large, both high and low prices adopted in the cyclic pricing strategy are larger than the static one.——这个结论呼应了Corollary 1在 small-size market 下的正确性,并且这时候不需要一定满足θ≥0.5\theta \geq 0.5θ≥0.5。
- 对比虚线(optimal static price,即 the welfare-maximizing price)发现:
- ?表明 L?L^*L? (the proportion of time that the high-price be charged, 高价阶段的时间占比 )随着 potential market size Λ\LambdaΛ 的增大而增大。即,市场越大,高价阶段时间占比越长。
4.3. An Alternative Information Scenario: Rating on the Net Utility
前面的rating考虑的是consumption utility:R?cwR - cwR?cw。naive customer只有在average rating 大于等于 price 的时候join。这种情况下,负的rating一定会赶走naive customer。
接下来讨论另一种场景,将price纳入考虑范围,即rating考虑的是 net utility:R?cw?piR-cw-p_iR?cw?pi?。这种场景下:
-
知道net utility & price:有些网站的提供的 average rating 会同时公布cost-effectiveness(e.g. price)信息,这时候潜在顾客可以对consumption utility进行推断;——这时候cyclic pricing strategy可以生效
-
只知道net utility:有些网站的提供的average rating score 只公布了 net utility,这时候naive customer 在 average rating 大于等于 0 的时候才 join。——这时候 static pricing strategy 就足够了,因为service provider 不能通过 cyclic pricing strategy 改变 naive customer 的行为。
the validity of the cyclic pricing strategy requires rating over both the net utility and the price。所以下面讨论上述中的情况1:
-
with the ratings on both the price and the net utility时:average rating 为负的时候,naive customer 依然可能会 join。下面用具体数值举个例子进行说明:
- At the beginning of a certain period, naive customers come and
they see a rating of average price $100 and an average rating of the net utility -10. 如果当前的价格为 $80,相当于平均价格20%的折扣。$20的折扣可以抵消-$10的负net utility,所以naive customer 会 join。 - 也就是说,with the knowledge about the average ratings on both the net utility and the price, incoming customers can still make their joining/balking decision by comparing the amount of price discount with the average rating over the net utility.
- 通过 average price p?\overline pp? + average rating of the net utility 推断 average consumption。如下, average rating on consumption utility η(PT)\eta(PT)η(PT) = average rating on net utility η′(PT)\eta'(PT)η′(PT) + average price p?\overline pp? :
- 然后通过比较推断的average consumption 和 current-period price 做出 joining/balking decisions。下面给出命题4:
- At the beginning of a certain period, naive customers come and
PROPOSITION 4. Give a pricing strategy PT, the joining-or-balking decisions of both types of customers remain the same under the two rating information scenarios, rating over the consumption utility only and rating over both the price and the net utility.
命题4表明,对于顾客来说,joining or balking 行为和之前的场景一样。对于provider来说,要基于顾客的行为做出 optimal pricing decision,所以也和之前的场景一样。即 4.1节和4.2节的结论在这里同样适用。
需要注意的是,naive customer 事前做出的 ex ante joining/balking decisions完全取决于 rating information,在前面的数值示例中,他们认为join会带来$10 的 net utility 收益,由于所有的naive customer都这么想,所以所有的naive customer都会加入,这就导致了服务系统变得overcrowded,结果所有的顾客最终在事后都获得了负的 ex post net utility。也就是说,由于naive customer忽略了同类型顾客的行为(peer-customers’ joining behavior),导致了ex ante and ex post payoffs 之间的差异。
—— ?This is actually the key difference between a goods market and a service market: quality in terms of congestion in a service system is endogenously (内生的)formed by customers themselves while quality in a goods market is exogenous(外生的).(商品市场和服务市场的主要区别:服务系统中拥塞对服务质量的影响是客户内生形成的,而商品市场中的质量是外生的。)
an ex ante “correct” decision may actually result in ex-post disappointing outcomes——所以本文的定价机制隐含了一个基本假设:这些naive customer 是 one-time(一次性)顾客。
? cyclic pricing strategy 为 service provider 带来更高利润的理由是,即使长期平均评级较低且为负,但是当 naive customer 观察到低于平均的价格时就能够获得额外收益。尽管消费后的他们并不满意还会给出很低的评分,但总体评分仍可以通过sophisticated customers发布的高评分来提升,这将会在下一个定价阶段吸引另一批naive customer。
5.Comparison between Profit Maximization and Welfare
Maximization
总结下前面的内容,我们讨论了2种情况下的service provider 的 optimal pricing strategies:1. welfare maximization;2. profit maximization。
-
当potential market size Λ≤Λ~\Lambda \leq \tilde \LambdaΛ≤Λ~ 时,profit-maximizing = welfare-maximizing;
-
当potential market size Λ≥Λ~\Lambda \geq \tilde \LambdaΛ≥Λ~ 时,welfare-maximizing 采取 static pricing,而profit-maximizing采取 cyclic pricing 。
下面来看下 cyclic pricing strategy 怎么影响 system’s welfare:
Figure 6 展示了随着potential marker size的变化, cyclic pricing strategy 造成的welfare loss。有意思的是:不是单调变化的!
- 当θ=0.4\theta = 0.4θ=0.4 时(蓝色),Λ\LambdaΛ 超过临界值Λ~=7.62\tilde \Lambda = 7.62Λ~=7.62时采取cyclic pricing strategy ;
- 在Λ~≤Λ≤Λ^=9.19\tilde \Lambda \leq \Lambda \leq \hat\Lambda=9.19Λ~≤Λ≤Λ^=9.19 范围内welfare loss 先增后减,在 Λ^=9.19\hat\Lambda=9.19Λ^=9.19 时减小到0。
- 在这个范围内,welfare maximization 要求只服务naive customer,socially optimal price is psw?=Vnp_{sw}^* = V_npsw??=Vn? ,因此maximal social welfare 为SW?=VnΛnSW^* = V_n \Lambda_nSW?=Vn?Λn?。
- 而在cyclic pricing strategy,high-price阶段只服务sophisticated customer,low-price阶段只服务naive customer,所以the difference in welfare only occurs during the high-price phase。
- 简言之,welfare loss 来自于 high-price 阶段只服务sophisticated customer导致underutilization of the system,因为此时sophisticated customers 的量比naive customer要少。
- 在Λ≥Λ^=9.19\Lambda \geq \hat\Lambda=9.19Λ≥Λ^=9.19 时,welfare loss递增。
- 在这个范围, welfare maximization 要求只服务sophisticated customer,
- 而cyclic pricing 在low-price阶段允许所有naive customer进入,所以 welfare loss 主要来自overutilization of the system during the low-price phase。 potential market size 越大,系统越拥挤。
- 在Λ~≤Λ≤Λ^=9.19\tilde \Lambda \leq \Lambda \leq \hat\Lambda=9.19Λ~≤Λ≤Λ^=9.19 范围内welfare loss 先增后减,在 Λ^=9.19\hat\Lambda=9.19Λ^=9.19 时减小到0。
- 当θ=0.5\theta = 0.5θ=0.5 时(红色) ,Λ\LambdaΛ 超过临界值Λ~=8.66\tilde \Lambda = 8.66Λ~=8.66 采取cyclic pricing strategy :
- welfare loss先减后增,在 Λ^=9.30\hat\Lambda=9.30Λ^=9.30 时减小到最低值。解释如下:
- θ≥0.5\theta \geq 0.5θ≥0.5时,socially optimal price is psw?=p(λb)p_{sw}^*=p(\lambda_b)psw??=p(λb?) (λb\lambda_bλb? is the socially desirable effective arrival) ,
- 而cyclic pricing cause underutilization in both phases.
welfare is the sum of the service provider’s profit and all customers’ utilities.The price becomes the internal transfer between the two parties, 所以social welfare 的表达式与price无关,而是由the number of customers served and each customer’s consumption utility 决定。
?The trade-off between the number of customers served and consumption utility per customer determines an optimal congestion level.
?welfare loss is mainly caused by the uneven workloads between the two phases—over-congestion in the low-price phase and underutilization in the high-price phase.(福利损失主要来自于两个阶段工作量的不平衡:低价阶段的过度拥挤和高价阶段的不完全利用)
但是,当the market size is not large 或者sophisticated customers comprise not less than half of the population时,cyclic pricing在两个阶段都会导致underutilization 而引起 welfare loss。
6. Conclusion
本文的研究问题:consider a typical service situation where customers are heterogeneous in information accesses: some customers know the service-related
information, whereas others do not; the latter relies on the buyer-generated information to make their queueing decisions. (两类客户,信息异质)
本文的模型方法: a high-low cyclic pricing strategy can be used to improve the profitability of a service provider without distorting customers’ ratings. (高-低价循环定价策略)
-
sophisticated customers join at high-price phase
-
naive customers join at the low price
-
During the high-price phase, the system is less congested, and thus the ratings are relatively high, which boosts the average rating and allows the provider to charge a price higher than naive customers’ expected consumption utility during the low-price phase.
模型有效性的前提条件:the potential market size to be above a threshold value such that congestion is a significant factor in affecting customers’ joining decision.(系统的拥堵程度是重要因素)
有趣的发现1:under the optimal cyclic pricing strategy, even though naive customers feel unsatisfied and post low ratings after consumption, the average rating can still be maintained by getting high ratings in the high-price phase. This, in turn, allows the service provider to obtain a higher profit than that under a static pricing strategy.(naive客户是被收割的韭菜)
有趣的发现2:although the cyclic pricing strategy can improve profitability, it harms social welfare. Welfare maximization prefers even workloads across periods, and hence prefers the static pricing strategy. However, the system can be either overor under-utilized under the cyclic pricing strategy, leading to welfare loss.(系统工作量的不平衡导致的社会福利降低)
本文研究的主要目的:show that even without manipulating customer ratings, a service provider can still take advantage of uninformed customers by implementing a cyclic pricing strategy and advertising the average rating to the incoming customers.(provider怎么在不操纵公开评级的条件下收割韭菜)
番外:From the viewpoint of naive customers, they shall take in the aggregated rating information with caution: when the products are service goods, an average rating in the past does not mean that they can get the same consumption utility if they join. How to protect naive customers’ welfare remains to be an interesting research question. (作为韭菜消费者们要注意,尽管服务很好,历史评级很高,你依然存在被收割的可能性)
—本文完—