Abstract
Which is better for a firm, job rotation or specialization, can be considered as an endogenously formed worker-indivisible job matching problem. We model this problem as a firm’s profit maximization problem under uncertainty, with and without overlapping generations. In both models, we show that among all possible job allocations, the rotation and specialization schemes are the only variations that can be optimal in terms of profits. Moreover, the rotation scheme is better when the productivity difference between post- and under-training workers is smaller, the uncertainty about job continuation in the future is more significant, or the slope of seniority wages is larger. The results indicate that firms in different environments prefer different worker-job matchings.
Similar content being viewed by others
Notes
As will be mentioned later, other related characteristics of the Japanese employment system are lifetime employment and seniority wages; those of the U.S. employment system are non-lifetime employment and non-seniority wages.
Evidence indicates that lifetime employment still remains in Japan, though it has recently become tenuous for some young cohorts. See Ito and Hoshi (2020) for a survey of recent evidence.
On a related note, Anderson (2012) constructed a model of a worker’s choice to be a specialist or generalist, and showed that it is rational to be a generalist when there are barriers to working on problems in other disciplines but problems are relatively simple.
Meyer (1994) analyzed the relationship between task assignments and promotion. Although we do not explicitly consider promotion in our model, we can take it into account by modifying our assumption regarding wages.
This implies that the event of one job loss is not independent of another, making the job assignment problem well-defined. Otherwise, there would be no room for job assignments.
This is the rule already explained in (4).
References
Anderson KA (2012) Specialists and generalists: equilibrium skill acquisition decision in problem-solving populations. J Econ Behav Organ 84:463–473
Antler Y (2015) Two-sided matching with endogenous preferences. Am Econ J Microecon 7:241–258
Aoki M, Okuno M (1996) Comparative institutional analysis of economic systems. University of Tokyo Press (in Japanese)
Bloch F, Cantala D (2013) Markovian assignment rules. Soc Choice Welf 40:1–25
Brünner T, Friebel G, Holden R, Prasad S (2021) Incentives to discover talent. J Law Econ Organ 38:309–344
Campion MA, Cheraskin L, Stevens MJ (1994) Career-related antecedents and the outcomes of job rotation. Acad Manag J 37:1518–1542
Carmichael HL, MacLeod WB (1993) Multiskilling, technical change and the Japanese firm. Econ J 103:142–160
Corbae D, Temzelides T, Wright R (2003) Directed matching and monetary exchange. Econometrica 71:731–756
Gale D, Shapley LS (1962) College admissions and the stability of marriage. Am Math Mon 69:9–15
Ito T, Hoshi T (2020) The Japanese economy, 2nd edn. The MIT Press
Iwai K (1999) Persons, things and corporations: the corporate personality controversy and comparative corporate governance. Am J Comp Law 47:583–632
Iwai K (2014) Capitalism and cultures: universality and particularity of the corporate system across societies. The Clarke Lecture, Cornell Law School
Kiyotaki N, Wright R (1989) On money as a medium of exchange. J Polit Econ 97:927–954
Konings J, Vanormelingen S (2015) The impact of training on productivity and wages: firm-level evidence. Rev Econ Stat 97:485–497
Kurino M (2014) House allocation with overlapping generations. Am Econ J Microecon 6:258–289
Li F, Tian C (2013) Directed search and job rotation. J Econ Theory 148:1268–1281
Meyer MA (1994) The dynamics of learning with team production: implications for task assignment. Q J Econ 109:1157–1184
Ortega J (2001) Job rotation as a learning mechanism. Manag Sci 47:1361–1370
Zwick T (2002) Continuous training and firm productivity in Germany. ZEW Discussion Paper No. 02-50
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We would like to thank an anonymous referee, Onur Kesten, Eiichi Miyagawa, and Suraj Prasad for their insightful comments and participants at the JEA Meeting for their comments. This research is supported by the JSPS KAKENHI Grant-in-Aid for Scientific Research (c) No. 16K03547 and Grant-in-Aid for Challenging Research (Pioneering) No. 20K20279, and by AMED under Grant Number JP21zf0127005. We have no relevant financial or non-financial interests to disclose. All remaining errors are our own.
Appendices
A Proof of Proposition 1
We complete the proof of Proposition 1 by checking the profit comparison in the following three cases.
Case 1: \(p>\frac{(1-\lambda )-2(w-w^{*})}{2\delta (2^{\alpha }-1-w)}\). We show that for each \(k\in \{1,\ldots ,8\}\), \(\pi (x^{R1})>\pi (x^{k})\). First, under Claim 1, \(\pi (x^{R1})>\pi (x^{S1})\). Second, from Table 2, for each \(k\in \{1,\ldots ,8\}\setminus \{2,7\}\), \(\pi (x^{R1})>\pi (x^{k})\). It remains to show the inequality for \(k\in \{2,7\}\).
where the inequality follows from \(p>\frac{(1-\lambda )-2(w-w^{*})}{2\delta (2^{\alpha }-1-w)}\). Let us define the function \(f:[0,1]\rightarrow \mathbb {R}\) by \(f(\lambda )=\lambda +\delta \left( \frac{1}{2}-\lambda ^{1/2}+\frac{\lambda }{2}\right) .\) Then, \(f(0)=\frac{\delta }{2}<1=f(1)\), \(f^{\prime }(\lambda )=1+\frac{\delta }{2}\left( 1-\frac{1}{\lambda ^{1/2}}\right) \), and \(f^{\prime \prime }(\lambda )=\frac{\delta }{4}\lambda ^{-3/2}>0\). Thus, function f is concave and there is a unique \(\lambda ^{*}\) such that \(f^{\prime }(\lambda ^{*})=0\), i.e., \(\lambda ^{*}=\left( \frac{\delta }{\delta +2}\right) ^{2}\in (0,1)\). Because our calculation gives us \(f(\lambda ^{*})=\frac{2\delta ^{2}+4\delta }{2(2+\delta )^{2}}>0,\) we can conclude that for any \(\lambda \in (0,1)\), \(f(\lambda )>0\). Thus, \(\pi (x^{R1})>\pi (x^{2})\) and \(\pi (x^{R1})>\pi (x^{7})\).
Case 2: \(p<\frac{(1-\lambda )-2(w-w^{*})}{2\delta (2^{\alpha }-1-w)}\). We show that for each \(k\in \{1,\ldots ,8\}\), \(\pi (x^{S1})>\pi (x^{k})\). First, under Claim 1, \(\pi (x^{S1})>\pi (x^{R1})\). Second, from Table 2, for each \(k\in \{1,\ldots ,8\}\setminus \{2,7\}\), \(\pi (x^{R1})>\pi (x^{k})\), and thus \(\pi (x^{S1})>\pi (x^{k})\). It remains to show the inequality for \(k\in \{2,7\}\).
where the inequality follows from \(p<\frac{(1-\lambda )-2(w-w^{*})}{2\delta (2^{\alpha }-1-w)}\). As shown in Case 1, for each \(\lambda \in (0,1)\), \(f(\lambda )>0\). Thus, under inequality (5), \(\pi (x^{S1})>\pi (x^{2})\) and \(\pi (x^{S1})>\pi (x^{7})\).
Case 3: \(p=\frac{(1-\lambda )-2(w-w^{*})}{2\delta (2^{\alpha }-1-w)}\). The last part of the proposition is now straightforward from the arguments in cases 1 and 2.\(\square \)
B Proof of Proposition 2
To prove Proposition 2, it remains to show lemmas 1 and 2. We start with some preliminaries.
1.1 B.1 Preliminaries
We need to calculate the value function depending on the states \(\omega ^{O}\) and \(\omega ^{A}\). Note that \(\omega ^{O}\) is the set of assignable jobs at some period, and \(\omega ^{A}\) is the set of available jobs at some period. For the simplicity of notation, when we write the state \((\omega ^{O},\omega ^{A})\), for example, we denote \((\omega ^{O},\omega ^{A})=(\emptyset ,A;A,B)\) for \((\{\emptyset ,A\};\{A,B\})\) where we omit the braces for sets. Thus, for value functions, we denote \(V(\omega ^{O},\omega ^{A})=V(\emptyset ,A;A,B)\) for \(V(\{\emptyset ,A\},\{A,B\})\). Because of the symmetric roles of jobs A and B, we only need to calculate the value function for the following five cases. (1) \(V(\emptyset ;A)=V(\emptyset ;B)=V(\emptyset ,B;A)=V(\emptyset ,A;B)\), (2) \(V(\emptyset ;A,B)\), (3) \(V(\emptyset ,A;A)=V(\emptyset ,B;B)=V(\emptyset ,A,B;B)=V(\emptyset ,A,B;A)\), (4) \(V(\emptyset ,A;A,B)=V(\emptyset ,B;A,B)\), and (5) \(V(\emptyset ,A,B;A,B)\).
1.1.1 B.1.1 Calculation of \(V(\emptyset ;A)=V(\emptyset ;B)=V(\emptyset ,B;A)=V(\emptyset ,A;B)\)
We calculate \(V(\emptyset ;A)\), which is obviously equal to \(V(\emptyset ;B)\), \(V(\emptyset ,B;A)\), and \(V(\emptyset ,A;B)\). We have the following three candidates for \(V(\emptyset ;A)\), depending on period-t allocations.
\(V^{11}\) | \(V^{12}\) | \(V^{13}\) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) |  | Alloc. at t | States at \(t+1\) |  | Alloc. at t | States at \(t+1\) | |||
 | \(s=1\) | \(s=2\) |  |  | \(s=1\) | \(s=2\) |  |  | \(s=1\) | \(s=2\) |  |
\(a^{t-1}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{A}=\{A\}\) | \(a^{t-1}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{A}=\{A\}\) | \(a^{t-1}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{A}=\{A\}\) |
\(a^{t}\) | \(A^{*}\) | A | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(a^{t}\) | \(\emptyset \) (\(A^{*}\)) | \(A^{*}\) (\(\emptyset \)) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(a^{t}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{O}=\{\emptyset \}\) |
Denote
Claim 2
\(V^{11}>V^{12}\) and \(V^{11}>V^{13}\). Thus, \(V(\emptyset ;A)=V^{11}\).
Proof
First, \(V^{11}-V^{12}=\pi (\emptyset ,A;A)-\pi (\emptyset ,\emptyset ;A)\). Thus, since \(\pi (\emptyset ,A;A)>\pi (\emptyset ,\emptyset ;A)\) according to Assumption 8, we have \(V^{11}>V^{12}\). We next show \(V^{11}>V^{13}\). Note that \(V^{11}-V^{13}=\{\pi (\emptyset ,A^{*};A)-\pi (\emptyset ,\emptyset ;A)\}+\{\pi (\emptyset ,A;A)-\pi (\emptyset ,\emptyset ;A)\}+\delta \{V(\emptyset ,A;A)-V(\emptyset ;A)\}.\) By Assumption 8, \(\pi (\emptyset ,A^{*};A)>\pi (\emptyset ,\emptyset ;A)\) and \(\pi (\emptyset ,A;A)>\pi (\emptyset ,\emptyset ;A)\). Also, as any allocation at state \((\emptyset ;A)\) is possible at state \((\emptyset ,A;A)\), we have \(V(\emptyset ,A;A)\ge V(\emptyset ;A)\). Thus, \(V^{11}>V^{13}\). \(\square \)
Thus, we have
Since \(V(\emptyset ,A;A)=\frac{\pi (A,A^{*};A)+\pi (A,A;A)}{1-\delta }\) (this will be shown in Section B.1.3),
1.1.2 B.1.2 Calculation of \(V(\emptyset ;A,B)\)
We have the following three candidates for \(V(\emptyset ;A,B)\), depending on period-t allocations.
\(V^{21}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(B^{*}\) | B | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) |
\(V^{22}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(A^{*}\) | A | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) |
\(V^{23}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(A^{*}\) | \(B^{*}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) |
Claim 3
\(V^{23}>V^{21}=V^{22}\). Thus, \(V(\emptyset ;A,B)=V^{23}\).
Proof
First, \(V^{21}=V^{22}\) by symmetry. Next, we have
The last inequality follows from the fact that \(w>w^{*}\), \(V(\emptyset ,A,B;A,B)\ge V(\emptyset ,A;A,B)\), and \(V(\emptyset ,B;B)\ge V(\emptyset ;B)\). \(\square \)
1.1.3 B.1.3 Calculation of \(V(\emptyset ,A;A)=V(\emptyset ,B;B)=V(\emptyset ,A,B;B)=V(\emptyset ,A,B;A)\)
We calculate \(V(\emptyset ,A;A)\), which is obviously equal to \(V(\emptyset ,B;B)\), \(V(\emptyset ,A,B;B)\), and \(V(\emptyset ,A,B;A)\). We have the following three candidates for \(V(\emptyset ,A;A)\), depending on period-t allocations.
\(V^{31}\) | Â | Â | Â | \(V^{32}\) | Â | Â | Â | \(V^{33}\) | Â | Â | Â |
---|---|---|---|---|---|---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) |  | Alloc. at t | States at \(t+1\) |  | Alloc. at t | States at \(t+1\) | |||
 | \(s=1\) | \(s=2\) |  |  | \(s=1\) | \(s=2\) |  |  | \(s=1\) | \(s=2\) |  |
\(a^{t-1}\) | A | A | \(\omega _{t+1}^{A}=\{A\}\) | \(a^{t-1}\) | A | A | \(\omega _{t+1}^{A}=\{A\}\) | \(a^{t-1}\) | A | A | \(\omega _{t+1}^{A}=\{A\}\) |
\(a^{t}\) | \(A^{*}\) | A | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(a^{t}\) | \(\emptyset \) (\(A^{*}\)) | \(A^{*}\) (\(\emptyset \)) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(a^{t}\) | \(\emptyset \) | \(\emptyset \) | \(\omega _{t+1}^{O}=\{\emptyset \}\) |
Claim 4
\(V^{31}>V^{32}\) and \(V^{31}>V^{33}\). Thus, \(V(\emptyset ,A;A)=V^{31}\).
Proof
First, \(V^{31}-V^{32}=\pi (A,A;A)-\pi (A,\emptyset ;A)\). Then, since \(\pi (A,A;A)>\pi (A,\emptyset ;A)\) under Assumption 8, we have \(V^{31}>V^{32}\). Next, we show \(V^{31}>V^{33}\). Note that \(V^{31}-V^{33}=\{\pi (A,A;A)-\pi (A,\emptyset ;A)\}+\{\pi (A,A^{*};A)-\pi (A,\emptyset ;A)\}+\delta \{V(\emptyset ,A;A)-V(\emptyset ;A)\}.\) Under Assumption 8, \(\pi (A,A;A)>\pi (A,\emptyset ;A)\) and \(\pi (A,A^{*};A)>\pi (A,\emptyset ;A)\). Moreover, \(V(\emptyset ,A;A)\ge V(\emptyset ;A)\). Hence, \(V^{31}>V^{33}\). \(\square \)
Thus, we have
1.1.4 B.1.4 Calculation of \(V(\emptyset ,A;A,B)=V(\emptyset ,B;A,B)\)
We calculate \(V(\emptyset ,A;A,B)\), which is obviously equal to \(V(\emptyset ,B;A,B)\). We have the following three candidates for \(V(\emptyset ,A;A,B)\), depending on period-t allocations.
\(V^{41}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | A | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(B^{*}\) | B | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) |
\(V^{41}=\pi (A,B^{*};A,B)+\pi (A,B;A,B)+(1-2p)\delta V(\emptyset ,B;A,B)+p\delta V(\emptyset ,B;A)+p\delta V(\emptyset ,B;B)\). Note that \(V^{41}\) is the value of the specialization.
\(V^{42}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | A | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(A^{*}\) | A | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) |
\(V^{42}=\pi (A,A^{*};A,B)+\pi (A,A;A,B)+(1-2p)\delta V(\emptyset ,A;A,B)+p\delta V(\emptyset ,A;A)+p\delta V(\emptyset ,A;B).\)
\(V^{43}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | A | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(A^{*}\) | \(B^{*}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) |
Claim 5
\(V^{41}>V^{42}\) and \(V^{43}>V^{42}\).
Proof
First,
By calculation, \(\pi (A,B^{*};A,B)>\pi (A,A^{*};A,B)\) and \(\pi (A,B;A,B)>\pi (A,A;A,B)\). Moreover, by symmetry, \(V(\emptyset ,B;A,B)=V(\emptyset ,A;A,B)\), \(V(\emptyset ,B;A)=V(\emptyset ,A;B)\), and \(V(\emptyset ,B;B)=V(\emptyset ,A;A)\). Thus \(V^{41}>V^{42}\).
We next show \(V^{43}>V^{42}\).
The last inequality follows from the fact that \(\pi (A,B^{*};A,B)-\pi (A,A;A,B)=(\lambda ^{1/2}-w-w^{*})+2w=\lambda ^{1/2}+w-w^{*}>0\), \(V(\emptyset ,B;B)\ge V(\emptyset ;B)\), and \(V(\emptyset ,A,B;A,B)\ge V(\emptyset ,A;A,B)\). Thus, \(V^{43}>V^{42}\). \(\square \)
At this stage, we cannot clearly say which is larger, \(V^{41}\) or \(V^{43}\).
1.1.5 B.1.5 Calculation of \(V(\emptyset ,A,B;A,B)\)
We have the following four candidates for \(V(\emptyset ,A,B;A,B)\), depending on period-t allocations.
-
Case 1: \((x_{t,1}^{t-1},x_{t,2}^{t-1})=(A,A)\). Then, the old worker is assigned job A though jobs A and B are assignable. Thus, the value candidate \(V^{51}\) in this case is equal to the value when only job A is assignable, i.e., \(V^{51}=V(\emptyset ,A;A,B)\).
-
Case 2: \((x_{t,1}^{t-1},x_{t,2}^{t-1})=(B,B)\). Then, similarly, the value candidate \(V^{52}\) in this case is \(V^{52}=V(\emptyset ,B;A,B)\).
-
Case 3: \((x_{t,1}^{t-1},x_{t,2}^{t-1})=(A,B)\). Denote the value candidate in this case by \(V^{53}\). Then, we have the following four subcandidates for \(V^{53}\), depending on the young worker’s assignment \((x_{t,1}^{t},x_{t,2}^{t})\).
$$\begin{aligned} V^{531}= & {} \pi (A,B^{*};A,B)+\pi (B,A^{*};A,B)\\{} & {} \quad +(1-2p)\delta V(\emptyset ,A,B;A,B)+p\delta V(\emptyset ,A,B;A)+p\delta V(\emptyset ,A,B;B). \end{aligned}$$Note that \(V^{531}\) is the value of the rotation. \(V^{532}=\pi (A,A^{*};A,B)+\pi (B,B^{*};\) \(A,B)+(1-2p)\delta V(\emptyset ,A,B;A,B)+p\delta V(\emptyset ,A,B;A)+p\delta V(\emptyset ,A,B;B).\) \(V^{533}=\pi (A,A^{*};A,B)+\pi (B,A;A,B)+(1-2p)\delta V(\emptyset ,A;A,B)+p\delta V(\emptyset ,A;A)\) \(+p\delta V(\emptyset ,A;B)\). \(V^{534}=V^{533}\) by symmetry.
\(V^{531}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | B | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(B^{*}\) | \(A^{*}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) |
\(V^{532}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | B | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(A^{*}\) | \(B^{*}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A,B\}\) |
\(V^{533}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | B | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(A^{*}\) | A | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,A\}\) |
\(V^{534}\) | Â | Â | Â | Â | Â |
---|---|---|---|---|---|
 | Alloc. at t | States at \(t+1\) | States at \(t+1\) | States at \(t+1\) | |
 | \(s=1\) | \(s=2\) |  |  |  |
\(a^{t-1}\) | A | B | \(\omega _{t+1}^{A}=\{A,B\}\) | \(\omega _{t+1}^{A}=\{A\}\) | \(\omega _{t+1}^{A}=\{B\}\) |
\(a^{t}\) | \(B^{*}\) | B | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) | \(\omega _{t+1}^{O}=\{\emptyset ,B\}\) |
Thus, \(V^{53}=\max \{V^{531},V^{532},V^{533},V^{534}\}\).
Claim 6
\(V^{531}>V^{532}.\)
Proof
This is because \(\pi (A,B^{*};A,B)>\pi (A,A^{*};A,B)\) and \(\pi (B,A^{*};A,B)>\pi (B,B^{*};A,B)\). \(\square \)
-
Case 4: \((x_{t,1}^{t-1},x_{t,2}^{t-1})=(B,A)\). Let \(V^{54}\) be the value in this case. Then, by symmetry, \(V^{54}=V^{53}.\)
1.2 B.2 Proof of Lemma 1: A necessary and almost sufficient condition for the rotation to be optimal
1.2.1 B.2.1 A necessary condition for the rotation to be optimal
Since our initial condition is \((\emptyset ,A,B;A,B)\), we focus on Section B.1.5 for \(V(\emptyset ,A,B;A,B)\). Suppose that the rotation is a profit-maximizing allocation. Then, the value is \(V(\emptyset ,A,B;A,B)=V^{531}\). In addition to the inequality of Claim 6, we have the following relations.
where \(V^{4}:=V(\emptyset ,A;A,B)=\max \{V^{41},V^{42},V^{43}\}\). Let
Then,
Thus, we have
Claim 7
\(V^{43}>V^{41}\Rightarrow V^{531}>V^{533}\).
To know the implications from (8) and (9) using (10) and (11), we explore the value of M. Since \(V(\emptyset ,A,B;A,B)=V^{531}\) and \(V(\emptyset ,A;A,B)=V^{4}\), we have
In the above M, we have the three unknowns: The first unknown is \(V^{531}\). It follows from Section B.1.5 that \(V^{531}=\pi (A,B^{*};A,B)+\pi (B,A^{*};A,B)+(1-2p)\delta V^{531}+p\delta V(\emptyset ,A;A)+p\delta V(\emptyset ,B;B)\). Thus, solving this equation for \(V^{531}\), we have
The second unknown is
The third unknown is \(V^{4}\) for which we have two cases from (10): \(M<1+w^{*}-w\) and \(M\ge 1+w^{*}-w\).
Case 1: \(M<1+w^{*}-w\). Then, \(V^{43}-V^{41}=M-(1-w+w^{*})<0\) and thus, \(V^{43}<V^{41}\). Thus, since \(V^{41}>V^{42}\) and \(V^{43}>V^{42}\) by Claim 5 in Section B.1.4, we have \(V^{4}=V^{41}\). Now, it follows from Section B.1.4 that \(V^{4}=V^{41}=\pi (A,B^{*};A,B)+\pi (A,B;A,B)+(1-2p)\delta V(\emptyset ,B;A,B)+p\delta V(\emptyset ,B;A)+p\delta V(\emptyset ,B;B).\) Since \(V(\emptyset ,B;A,B)=V^{4}\) and \(V(\emptyset ,B;A)=V(\emptyset ;A)\), this becomes
Thus, we have obtained the three unknowns. Hence, \((V^{531}-V^{4})\) in M can be calculated as
Thus, we can calculate the value of M as follows.
Therefore, we substitute this M into (11) to get the following implication from (8).
On the other hand, we use (15) to get the following implication from (9).
Therefore, for Case 1, we have the two necessary conditions (16) and (17). In these equations, since \(-\lambda ^{\frac{1}{2}}+(1-2p)\delta \lambda ^{\frac{1}{2}}<0\), (17) implies (16). Hence, we have (17) as a necessary condition.
Case 2: \(1+w^{*}-w\le M\). Then, \(V^{43}-V^{41}=M-(1-w+w^{*})\ge 0\) and thus, \(V^{43}\ge V^{41}\). Thus, since \(V^{41}>V^{42}\) and \(V^{43}>V^{42}\) under Claim 5 in Section B.1.4, we have \(V^{4}=V^{43}\). Now, it follows from Section B.1.4 that
Then, (9) is automatically satisfied as follows.
Thus, we substitute this value \((V^{531}-V^{4})\) into (12) to get \(M=(1-2p)\delta \lambda ^{\frac{1}{2}}+p\delta (V(\emptyset ,A;A)-V(\emptyset ;A)).\) We substitute this M into (11) to get the following implication of (8).
Therefore, for Case 2, we have the necessary condition (18).
In sum, we have (17) for Case 1 or (18) for Case 2. Since (17) implies (18), we have (17) as a necessary condition for the rotation to be optimal. Hence, \(p\ge \frac{1-w+w^{*}-\lambda ^{\frac{1}{2}}}{\delta \{(1+\lambda )^{\alpha }+2^{\alpha }-\lambda ^{\alpha }-1-2w\}}\).
1.2.2 B.2.2 An almost sufficient condition for the rotation to be optimal
Suppose that
The value of the rotation, \(\pi ^{R}\), is
We need to show
Here, the last three inequalities follow from the fact that \(V^{531}\ge V^{51}\) and \(V^{51}=V^{52}=V(\emptyset ,A;A,B)\equiv V^{4}=\max \{V^{41},V^{42},V^{43}\}\).
Claim 8
The three inequalities (21), (25), and (27) are sufficient for all of the above inequalities.
Proof
Suppose that (21), (25), and (27) hold. First of all, Claim 6 implies (20). Note that when (25) is true, by Claim 5 (\(V^{41}>V^{42}\)), (26) holds. Hence, it follows from (25), (26), and (27) that \(V^{531}\ge V^{4}\). This implies that (22) and (23) are true. On the other hand, since we have (20) and (21), we have \(V^{531}=V^{53}\) and thus (24). \(\square \)
From now on, we will check (21), (25), and (27).
-
Check whether \(V^{531}\ge V^{533}\), or (21).
-
Check whether \(V^{531}\ge V^{41}\), or (25).
$$\begin{aligned} V^{531}-V^{41}= & {} \pi (B,A^{*};A,B)-\pi (A,B;A,B)\\{} & {} +(1-2p)\delta \{V(\emptyset ,A,B;A,B)-V(\emptyset ,B;A,B)\}+p\delta \{V(\emptyset ,A;A)-V(\emptyset ;A)\}\\\ge & {} \pi (B,A^{*};A,B)-\pi (A,B;A,B)\\{} & {} +p\delta \{V(\emptyset ,A;A)-V(\emptyset ;A)\}\quad (\because V(\emptyset ,A,B;A,B)\ge V(\emptyset ,B;A,B))\\> & {} \pi (B,A^{*};A,B)-\pi (A,B;A,B)+1-w+w^{*}-\lambda ^{\frac{1}{2}}\quad (\because ((19)))\\= & {} \lambda ^{\frac{1}{2}}-w-w^{*}-(1-2w)+1-w+w^{*}-\lambda ^{\frac{1}{2}}=0. \end{aligned}$$ -
Check whether \(V^{531}\ge V^{43}\), or (27).
$$\begin{aligned} V^{531}-V^{43}= & {} \pi (B,A^{*};A,B)-\pi (A,A^{*};A,B)\\ {}= & {} \lambda ^{\frac{1}{2}}-w-w^{*}-(0-w-w^{*})=\lambda ^{\frac{1}{2}}>0. \end{aligned}$$
1.3 B.3 Proof of Lemma 2
Suppose that the rotation is not optimal. Then, we will check the two cases: \(V^{533}>V^{531}\) and \(V^{533}\le V^{531}\).
Case 1: \(V^{533}>V^{531}\). Since \(V^{531}>V^{532}\) under Claim 6, we have \(V^{53}=V^{533}\).
Claim 9
\(V^{41}>V^{43}\) and thus \(V^{4}=V^{41}\).
Proof
Since \(V^{533}>V^{531}\) holds, we have
Thus,
Thus, \(V^{41}>V^{43}\). Moreover, since \(V^{41}>V^{42}\) under Claim 5, we have \(V^{4}=V^{41}\). \(\square \)
Claim 10
\(V^{51}>V^{533}=V^{53}\) and thus \(V^{5}=V^{51}\).
Proof
We have
Thus, \(V^{51}>V^{533}\). Then, since we know \(V^{533}=V^{53}\) and \(V^{51}=V^{52}\), we have \(V^{5}=V^{51}\). \(\square \)
Note that we have \(V^{51}=V^{4}\), \(V^{5}=V^{51}\) (\(\because \) Claim 10), and \(V^{4}=V^{41}\) (\(\because \) Claim 9). Thus, \(V^{5}=V^{41}\), which is the specialization value. This means that the specialization is optimal.
Case 2: \(V^{533}\le V^{531}\). Then, since \(V^{531}>V^{532}\) under Claim 6, we have \(V^{53}=V^{531}\), which is the rotation value.
Claim 11
\(V^{51}=V^{4}>V^{531}\).
Proof
By definition, \(V^{51}=V^{4}\). Suppose to the contrary that \(V^{51}=V^{4}\le V^{531}\). Then, \(V^{531}\ge V^{4}=V^{51}=V^{52}\), which means that the rotation is optimal. However, this contradicts the initial argument that the rotation is not optimal. \(\square \)
Claim 12
\(V^{41}\ge V^{43}\) and thus \(V^{4}=V^{41}\).
Proof
Suppose to the contrary that \(V^{41}<V^{43}\). Then, since \(V^{42}<V^{41}\) (\(\because \) Claim 5), we have \(V^{4}=V^{43}\). Thus, by Claim 11, \(V^{43}>V^{531}\). However, this inequality contradicts the following.
Hence, we have \(V^{41}\ge V^{43}\). Moreover, as \(V^{41}>V^{42}\) (\(\because \) Claim 5), we have \(V^{4}=V^{41}\). \(\square \)
Under claims 11 and 12, we have \(V^{51}=V^{4}=V^{41}>V^{531}=V^{53}\). This means that the specialization is optimal.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Kurino, M., Kurokawa, Y. Job rotation or specialization? A dynamic matching model analysis. Rev Econ Design (2023). https://doi.org/10.1007/s10058-023-00345-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10058-023-00345-7