The Economic Journal, 96 {December 1986), 961-974 Printed in Great Britain

TRANSFERABLE SKILLS, JOB MATCHING, AND THE INEFFICIENCY OF THE 'NATURAL' RATE OF UNEMPLOYMENT*
Ben LockvDood The idea of the ' natural' rate of unemployment has been with us since Friedman (1968). This can be interpreted simply as an assertion of the long-run homogeneity of the Walrasian or flexible-price IS-LM system in prices (see e.g. Sargent (1979)5 P- 325) Over the last ten years, however, attempts have been made to model the ' natural' rate explicitly, following the original Friedman/Phelps idea that it could be thought of as the equilibrium rate of unemployment in a search framework. Although the main focus of attention, especially on the empirical side, has been on the determinants of the equilibrium rate and changes in this rate (see Johnson and Layard, 1984) it was soon recognised that in a search environment the equilibrium rate may not be efficient. This point was made by Tobin (1972), among others, who said:' The external effects are the familiar ones of congestion theory. A worker deciding to join a queue or stay in one considers the probabilities of getting a job, but not the effects of his decision on the probabilities that others face... external effects also occur in the decisions of employers whether to fill a vacancy with the applicant at hand or to wait for someone more qualified.' Recent papers by Diamond and Pissarides (see Diamond, 1981, 1982; Pissarides, 1983, 1984) have attempted to model these external effects in a stochastic matching model that can be described briefiy as follows. There are a number of jobs and workers, who are paired by a stochastic matching process. The number of jobs (or workers) may be fixed or determined by free-entry conditions. Once a job and worker are paired, the productivity of the match is determined by random draw, and the partners must decide whether to accept or reject it. There are potentially three margins of decision-making in this type of model; the decision to enter (or exit) the market, the choice of lowest acceptable (reservation) productivity, and the choice of search or advertising intensity. It has been shown that individual decision-making at each of these three margins may be socially inefficient. What these inefficiencies have in common, as Pissarides (1983) makes clear, is that they are all due to external effects that work through the matching technology; that is, when individuals choose to enter a market, accept a match, or search more intensively, they ignore the impact of their decisions on the matching probabilities of other agents in the market. One important limitation of the Diamond-Pissarides model is that both firms and workers are assumed homogeneous. We show in this paper that if this
* I would like to thank an anonymous referee whose comments have greatly improved both the exposition and substance of the paper. [961I

962

THE ECONOMIC JOURNAL

[DECEMBER

assutnption is dropped, the matching equilibrium exhibits a new kind of inefficiency, due to external effects that work through match acceptance probabilities, rather than matching probabilities. The distinction between the two is that the latter is the probability that a worker encounters a firm, or vice versa, within a given time period, and the former is the probability that having met, the two parties will decide to produce rather than continue to search for new partners. In particular, we show that if workers differ in inherent skills, then under a wide range of conditions the presence of high-skill workers lowers the equilibrium acceptance probabilities of low-skill workers below what is socially desirable. Indeed, if the variation in the match-specific component of productivity is low enough, this externality implies that aggregate employment and output will be too low in equilibrium. It is important to note that this external effect will occur irrespective of whether the matching probabilities are fixed or endogenous.^ Furthermore, the cause of this inefficiency is transparently simple, especially when there is no match-specific component of productivity - i.e. the output of the match is entirely determined by the skill or productivity ofthe worker - and when matching probabilities are fixed. For then there is no social gain to rejecting a match with any worker, no matter how low his skill - he is not 'crowding out' a high-skill one. From the point of view ofthe individual firm, on the other hand, with only one job to fill, it may be better to reject a low-skill worker in the hope of being matched with a better one. Therefore, as long as the firm has a say in the choice of reservation productivity (which will be the case as long as the worker does not appropriate all the surplus from the match), equilibrium may be inefficient. Interestingly, as this inefficiency can take the form of re-employment probabilities being more dispersed than is socially desirable, the usual equity/efficiency trade-off is broken - there are policies which will both improve efficiency and equalise the distribution of re-employment probabilities among workers. One such policy is a raising of worker bargaining power. Others are profit and wage taxes, although the effect of these taxes is not straightforward because they can be partially shifted by a change in equilibrium worker bargaining power. We examine the effect of such taxes in some detail below; the main conclusion is that a profit tax, (unless it is 100%) needs to be accompanied by an employment subsidy to the firm in order to be effective in reducing inefficiency, but that a wage tax can achieve an efficiency improvement on its own. The plan of the paper is as follows. The model is laid out in Section I, and efficient match acceptance rules are derived in Section II. In Section III the equilibrium matching rules are derived, and are compared in some detail with the efficient ones. Section IV discusses the role of taxes in restoring efficiency, and we conclude by discussing some possible extensions of the analysis to the cases where firms are heterogeneous, or where worker productivities are not directly observable.
1 The conditions under which matching probabilities are exogenous are that (a) the numbers ofjobs and workers are always equal to one another; (A) the matching process exhibits constant returns to scale, and (c) search and advertising intensities cannot be varied. We make these assumptions throughout the paper in order to present our inefficiency result as clearly as possible.

1986]

'NATURAL' RATE OF UNEMPLOYMENT

963

I. THE MODEL

The model is a generalisation ofthe one set out in Pissarides (1984). The economy is composed of workers and jobs. Firms are identified with jobs. There are L workers who are independently drawn from a discrete distribution of abilities aj,..., a^,fl^> o, with probabilities 7ri... n^, and J identical jobs. A worker of ability a^ when matched with a job can produce y = d + a^. Both 6 and a^ are common knowledge. The match-specific productivity element, 6, is distributed continuously on [0, 00) with density function/(^) > o. This assumption allows output to depend both on a match-specific component, 6, and the transferable skill of the worker, a, and is the distinctive feature of our analysis. Jobs and workers are brought together by a matching technology in which the number of matches in a given unit time period, X{ U, V) is a function of unemployment and vacancy numbers (see Diamond, 1982 for a full discussion). To abstract from the externalities arising from match acceptance or rejection investigated by Pissarides, we assume that the number of jobs is equal to the number of workers, and that X exhibits constant returns to scale. It should be noted that the first condition requires efifectively that the number of jobs and workers is fixed. While this may not be an unreasonable assumption in the case of workers, it has been argued (see Pissarides 1983, 1985a) that job numbers are better thought of as endogenous, perhaps being determined by a free-entry (i.e. zero profit) condition. Finally, matches break up at an exogenous rate s per period. We shall only be concerned with steady states in the model, i.e. where flows into and out of unemployment are equal for each ability type. The expected total flow into unemployment for each type is (Z^ - t/j) s, or in per capita terms (i u^)s, with M j . = UilL^. Tbe flow out of unemployment for any type is calculated as follows. The probability of any unemployed worker making a job contact is, in fact, constant: X{U,V)/U = X{i,i) = x by constant returns to scale and U = V. The probability of such a match being accepted is q.^. There are f/,- such workers, so the total expected flow is ^^ xL^, or in per capita terms qiXU^. By the law of large numbers, actual and expected flows per capita can be taken to be equal, and so we can write the flow equilibrium conditions
z = i,2, ...,n. (i)

Solving for the equilibrium unemployment levels, we have:

i = i,2,...,n. (2)

II. EFFICIENT JOB-MATCHING RULES

We are now in a position to derive the eflicient, or socially optimal, rules for match formation. As both firms and workers are assumed to be risk-neutral and tbere are no vacancy costs or unemployment costs and benefits, the appropriate maximand is the discounted flow of output per capita, which by the law of large

964

THE ECONOMIC JOURNAL

[DECEMBER

numbers is non-random. There is no real loss of generality in taking the maximand to the steady-state flow of output, as in what follows steady-state rules are either identical to the rules which maximise discounted flows, or correspond to a special case of the latter where the interest rate is set to zero.^ First, observe that from the law of large numbers, output per capita can be written

Y=i,n,{i-Ui)YI where a,, is as in (2), and

Yt = a,+ rdf{d)dd/[i-F{d,)].

(3)
(4)

Now, from (2), the unemployment levels u^ are determined by the q^. In turn, each g^ is determined by a critical reservation index of ability k, and job-speciflc productivity, 6^, such that the match is rejected ifi < k or 6 <6i. In other words, qf = x[i -F{6i)] if t > k, and ?< = 0 otherwise. We show in this section that the efficient reservation values (i.e. those that maximise the steady-state flow of output) have two properties. First, matches should be formed whatever the ability of the worker (i.e. k = i) and second, the reservation job-specific productivity 6i depends non-trivially on i. It is immediately obvious from (3) that k = 1 maximises the flow of output, as long as P > o, and the latter holds by the assumption that the a^ and 6 are nonnegative. In other words, no match should be rejected solely because the worker's ability is too low.^ Next, recalling the definition of w ^ from (2), and using the fact that

if the 6^ maximises r i t follows that each 6^ maximises

on o ^ Of <QC. It can be shown that this problem always has a solution.' If there is an interior solution, then 6^ satisfies [

where

d\= {'^

When a = o, (6) reduces to formula (29) in Pissarides with b = k = o, using

1 More precisely, the rule that all ability types should be matched is also true with discounting, and the formula for the optimal dj in (6) is only modified by the addition ofan interest-rate term in the denominator of the right-hand side of (6). This can be shown using the method of Diamond (1980). * Note that this depends crucially on there being at least as many jobs as workers. If there were more workers than jobs, then the rejection of a low-skill worker would raise the matching probabilities of all types of workers, including the highly skilled, and this may be enough for rejection of a low-skill match to be efficient. This follows from the fact that as 0 -* 00, (5) goes to zero, whereas at 0< = o, it is greater than zero.

1986]

' N A T U R A L ' RATE OF UNEMPLOYMENT

965

the fact that Z ( i , i) s dX/du-\-dX/dv from Euler's theorem. In addition, ^^ is strictly decreasing with a as long as ^^ > o, as from (6),
ddj ^

-s

da ~ s + [i It is worth remarking on the fact that ^^ depends negatively on a,- but not on the other aj,j 5^ i. The reason why it is independent of a,- is simply that there is no social gain from rejecting a worker of skill z, just in case a worker of higher skill j comes along. Rather, a worker of skill i should be rejected only when his jobspecific productivity is too low, so he might contribute to social output more by going elsewhere. But going elsewhere takes time, and the forgone output during this time is directly related to a^. So a higher a,- makes the search cost higher and reduces the reservation 6.

III. EQUILIBRIUM

In this section we describe the equilibrium of the transferable skills model and explain how and why it is inefficient. Any equilibrium is fully described by a wage function w{d,i) which specifies the wage to be paid to a type i worker in a match with match-specific productivity 6, and reservation job-specific productivities 6^, and an ability k such that a job match is rejected if and only liO < d^ or i < k. In turn, these reservation values will be different from the socially optimal ones - that is the main point of the paper - but in what follows we will usually be discussing equilibrium values, and so will not explicitly superscript them as such, except when the need arises. What we will show is that if either (i) 6 is not 'very' random, or (ii) workers have little bargaining power, then more highly skilled workers impose a negative externality on the less skilled by lowering re-employment probabilities of the latter below the socially optimal level. In the case that 6 is non-random, in fact, it is possible to show that some low-skill workers will never be employed in equilibrium, implying also that aggregate output and employment are below efficient levels. This is not necessarily the case if 5 is highly random; in fact, it is possible to show that whatever the distribution of ^, if worker bargaining power is low enough then the re-employment probabilities of the highly skilled are at least as high as the efficient levels, and will usually be strictly above the latter if 6 is random, and so the possibility that this more than offsets the under-employment of low-skill types cannot be ruled out. We turn now to a description of the equilibrium. The equilibrium wage and reservation values are determined by asset equations implicitly defining net worths, or asset values, for both firms and workers, and a rule for wage determination. Let Wg{6,i) be the net worth of an employed worker of ability a^ in a match with productivity 6, and M^(i) be the net worth of an unemployed worker of type i. Similarly, let lVf(6, i) be the net worth of a job with a match of productivity di + d, and W^ the net worth of a job vacancy. These values must satisfy the standard asset equations that the rate of return

966

THE ECONOMIC JOURNAL

[DECEMBER

times net worth equals flow income plus expected capital gain. For the workers, these asset equations are: (7) (8) with qwi = [i P{&i)]} aiid where (9) lVl{i) is in fact the expected net worth from employment of a type i worker, conditional upon a reservation productivity 5^. Analogous equations for jobs as assets are
)]

(10)
(11)

where

and where n* is the probability that the worker matched with the firm is of type i, and is in fact an endogenous variable, being n^ times the unemployment rate of type i workers, appropriately normalised:
^i n = UiTTj 2 UiTTf.
= 1

(13)

The reservation 6^ are now determined by the condition that each side of the market must be indifferent between rejecting and accepting the match, given that the worker is of type i, i.e.
i)=W^

(14)

The precise definition of the reservation ability index k is given below. Although at first sight it is possible that (14) and (15) give different answers, the rule we adopt for wage determination ensures that there is no inconsistency. Following Diamond and Pissarides, we suppose that the total surplus from the match, W^{6,i) +Wf{9,i) -[W,. + lii{i)] is divided up according to a generalised Nash bargain, so that the worker receives a fraction fi of it.^ Then the wage is implicitly determined by

It should be clear from (16) that (14) and (15) are in fact identical equations.
' Such a rule can be justified as the outcome of an explicit bargaining process where matched pairs then bargain over the division of the surplus from the match by making alternating offers, until one offer is accepted. (See Rubenstein and Wolinsky, 1984, or Binmore and Herrero, 1984.) The distinctive feature of these models that distinguishes them from simple bilateral bargaining models is that during bargaining both workers and firms continue to search for better matches. When a bargain is struck, both partners cease searching and production begins. The division of the surplus is given by a formula for fi which involves the matching probabilities as well as the interest rate, i.e.

1986]

' N A T U R A L ' RATE OF UNEMPLOYMENT

967

To bring out the issues clearly, we first examine the special case where there is no match-specific element to productivity. We show that for this case, matches involving low-ability types may be rejected, implying that the equilibrium is inefficient, and output and employment are too low relative to the social optimum. (a) A Special Case: d so In this case, q.^ = i if j is an ability type that is employed in equilibrium, and qy,i = o otherwise. The first step is to obtain an equation for the wage from equations (7)-(i3) and (16): w{i) = rWJi) +fi[a,-rhi{i) -rPfJ, i = k,...,n. (17) This is a standard equation for such matching models (see e.g. Pissarides, 1984, equation (13)) and its interpretation is that the wage must yield the worker the flow benefit from unemployment plus fraction ^ of the surplus from the match. Next, we can calculate the fiow benefits from unemployment and vacancy states explicitly:

""W-r
and with
i=k i=k

=* "
r + s + q^x

(-8)

?/ = 2; rrt a^=i n^aj ^ n*, ;" = S nf wjx nf,

I i=k i=k I i=k

S O that qf is the probability that the firm will hire a worker that it is matched with, a* is the average productivity of a random draw from the pool of workers, and finally w^ is the average wage. Note the asymmetry between (18) and (19), which does not arise in the Diamond-Pissarides type of model; namely that the return to unemployment for a worker depends only on his own wage, but the return of a vacancy to a firm depends on the average productivity minus the average wage. Substituting (18) and (19) into (17), and using the asset equations, we can solve explicitly for the expected wage and the individual wages in terms of the parameters: ; = A*fl, (20) with and w{i) with /?** =
This formula poses no problem in the case where 6 is degenerate, as then q^i is either o or i, and so the fii are constant across all types that are ever employed. When 0 is non-degenerate, however, this need not be the case, and so we must conclude that a rigorous derivation of the fi^ implies that they may differ across types, which greatly complicates the analysis, but does not fundamentally change the efficiency result. For this reason we assume throughout that fi is constant across types.
29 ECS 96

968

THE ECONOMIC JOURNAL

[DECEMBER

In Other words, the wage is proportional to the productivity ofthe worker minus a term which, from (19) and (20), is the return to a vacancy for the firm. Note that as long as ^ < i, the wage of type i worker depends on the productivities of other types in the economy, a phenomenon not apparent in the DiamondPissarides model. It is apparent from (17) that the critical index k is the smallest index k such that the surplus from a match, a^ rH{,{k) rW^, is non-negative. (It is easily checked that this is consistent with (14) and (15) above holding.) Now, from (18), rM^(t) is proportional to w{i), so from (17), the non-negative surplus condition is equivalent to iu{i) ^ 0. Thus, from {21), k is the smallest index such that
.
(22)

The first thing to note is that if fi is small (i.e. worker bargaining power is low) A : > I, so that the equihbrium will be efficient, as claimed. It is also apparent that as /?->i, eventually A ; = i, so that if worker bargaining power is high enough, the equilibrium will be efficient. The precise explanation for this result is as follows. There is a fundamental asymmetry in the job-matching rules that firms and workers would follow if one or the other could appropriate all the surplus. If the worker appropriates all the surplus (i.e. /? = i), then when considering whether to accept or reject a job he will compare his current gain, a^, with the expected discounted gain that he could get by rejecting the match, which will be some fraction of a^, and so he will always accept the match. A firm, by contrast, considers other workers' productivities too - in fact, what the firm does is to compare the productivity of a prospective match, a^, with the average productivity of another 'draw' from the worker population a. If the former is low enough relative to the latter, the match is rejected. Therefore, decisions will be inefficient if firms have a say in job acceptance, i.e. in all cases where fi < i. There is one problem with this equilibrium, however - workers of types below k, the critical index, will never be employed. Hence for these types there is no return to search, and so if there were even the slightest cost to search they would drop out of the search process, which would have an impact on the matching probabilities of the firms and remaining workers. One could model this 'discouraged worker' effect by endogenising the matching probabilities. This would, however, be rather complicated, while the basic inefficiency result would not change; that is, the equilibrium would now have low-productivity types quitting the market, which would not be socially optimal as long as search costs were low enough. We now turn to the case where the match-specific productivity is genuinely random.
{b) The General Case; d Random Again, the first step is to write the wage zs'm (17): W,], for i = k,...,n. (23)

1986]

' N A T U R A L ' RATE OF UNEMPLOYMENT

969

In this case, to save time, we proceed by writing the returns to unemployment and vacancy states explicitly in terms of the parameters:

and

and where c^j is the ijth element of C = (I-l-r)-ir and d^j is the ijth element of D = I - (I + r ) - T . Finally, the elements of T are given by

(26)
,
\

i=k

and

Now, it is apparent from (22) that the reservation < ? < are the ones for which there is no surplus from the match, or such that flf+ (9^ = rW^{i) +rW^. Writing this out in full, using (24) and (25), we require that the 6^ satisfy tht n-k + i simultaneous equations a, + ^. =

xq^i S cM + Of) xZnf q^i \ 2 <-,(

1* + J z ^ b ^
i=k

and the critical index k is the smallest index for which all these equations have solutions with all ^^ < 00 (or d^ < d if ^ has bounded support). Comparing (28) with (6), it is apparent that the equilibrium 6^ will be inefficient, but in general it is difficult to say anything about the direction of the inefficiency. However, it is possible to obtain results for the cases when worker bargaining power is either very high or very low. First, as ^-> I, all the elements of T go to +00, implying that C goes to the identity matrix, and D to the null matrix. Now if ^Z^,. = o, and c^^ 1, c^j = o, j ^ I, ec[uations (28) are identical to (6), so that our earlier result that the equilibrium becomes efficient as fi goes to i is confirmed. The other case is where fi^o. Here, all the elements of T go to o, so C goes to the null matrix and D to the identity matrix. In this case, equations (28) become:

'Ln';[i-F{dj)]{aj + dl)

(29)
29-2

970

THE ECONOMIC JOURNAL

[DECEMBER

Note first that if there is only one ability type, (29) reduces to (6), confirming Pissarides' result that the equilibrium will be efficient. However, we can now say something about the direction of the inefficiency with several types. Let ^| and df be the socially efficient and equilibrium reservation values respectively, where the ^f solve (29).
Proposition

There exists &nl,k <l <n, such that for i > I, OT ^ <9|, and for i ^ I, df ^ 6%. Furthermore, at least one of each of the inequalities is strict. The proof of this is to be found in the appendix to this paper. This result, combined with the preceding argument, says that when fi is low enough the equilibrium distribution of reservation values (and hence re-employment probabilities) across types is more dispersed than is socially optimal; high-skill types enjoy higher re-employment probabilities than they should, at the expense of low-skill types. This has the striking implication that the usual equity/efficiency trade-off is broken - there are policies which equalise rates of unemployment between skill types and which also increase efficiency. We investigate such policies in greater detail in the next section.

IV. TAXES, WORKER BARGAINING POWER AND EFFICIENCY

From the arguments of the preceding section it is clear that for the case where d is not very variable the degree of inefficiency of equilibrium turns on (a) the relationship between the profit to the firm of employing a given worker, a^ - w^, and the expected profit it can get from another random draw, a^ w^, and {b) the degree of worker bargaining power, fi. We have already remarked that one way of achieving efficiency of equilibrium is to raise worker bargaining power so that workers appropriate all, or nearly all, of the surplus. Another way of achieving complete efficiency is to impose a 100 % profit tax, combined with lump-sum redistribution of the proceeds, as this would equalise the firm's aftertax profits from each type at zero, and it would therefore be indifferent about whom it hired. One might conjecture from this that introducing a profit tax of less than IOO % will improve efficiency from the no-tax status quo. It is also of interest to know more generally how, in equilibrium, taxes interact with worker bargaining power to affect efficiency. It turns out that these two questions are closely related. An increase in a profit tax is (partially) shifted to workers by a lowering of the workers' share of the average product, fi* (recall that ; = fi* of), and the tax rate affects the reservation productivity only through fi*, as we show below. Therefore, we have the striking result that the imposition of a less than 100% profit tax decreases efficiency, whereas a 100 % tax will restore full efficiency that is, there is a discontinuity of the equilibrium matching rules in the tax rate at this point.

1986]

' N A T U R A L ' RATE OF UNEMPLOYMENT

971

The introduction of a wage tax will also have perverse effects, for the same reason; some ofthe tax will be shifted on to the firm by an increase in effective worker bargaining power, ^*, so the efficiency of equilibrium will rise. Finally, it turns out that if the proceeds from the tax are distributed to either firms or workers not in a lump-sum fashion, but in a way that varies with the state that they occupy, then this discontinuity may be eliminated. For example, if some positive fraction ofthe proceeds is redistributed only to firms who are currently employing a worker (i.e. a firm employment subsidy) or only to workers that are currently employed (a worker employment subsidy) then equilibrium becomes efficient as the profit or wage tax rate goes to unity. It is worth noting that such discontinuity of equilibrium in the tax parameters does not arise in Walrasian models, except in pathological cases. Our results would seem to indicate that such discontinuities are the norm in matching models, and so one should be cautious in applying arguments derived in a Walrasian setting to these models. We now turn to a formal analysis of the effects of profit and wage taxes. Let Tf be the profit tax rate, so the firms' net profit from employing a type i worker is [iTf) {a^-w^), and T^ the wage tax rate, so that the a!fter-tax wage is {i~Tj)Wi. Then it is simple to show that if the proceeds from the taxes are distributed in a lump-sum fashion, (22) becomes

r + s + xqf

'

where

First, from (30), it is apparent that for fi strictly between o and i, fi*{Tf,Tj) is decreasing in T^ and increasing in T^. The reason for this is quite simple. For a given profit tax rate (i fi*) {i -T^) is the firm's equilibrium post-tax share of the average product ofthe worker, a. As i - T^ falls with an increase in the profit tax rate (i -fi*) rises to compensate the firm for this, so fi* must fall. An exactly similar argument explains why fi* is increasing in T^. Next, it is possible to show^ that the index k is non-increasing in fi (we have already argued intuitively that this is the case in the previous section). Therefore, by the previous argument, k will be non-increasing in a wage tax, and non-decreasing in a profit tax, so that the former will improve, and the latter worsen, efficiency. Now, suppose that some of the tax revenue is redistributed as lump-sum payments to firms whose vacancies are filled, and also lump-sum payments to workers who are currently employed. Let these payments be by and A^ respectively. Then the net profit from a filled vacancy is (i - T^) (a^ - w^) + bj, and the
The proof is as follows. Let the right-hand side of (22') as a function of/? and k be g{p, k). We know that it is decreasing in p. We know also that from the definition of k, (a) "i 5 = g{P, i) I > k, and (A) a^_^ < g{fi, k- 1). Now suppose that /J' < /S but k' < k. Then from (a), at_, ^ g{fi', k- i) > g{fi, k- i) which contradicts (A). Hence fi' ^ fi implies k' ^ k after all.

972

THE ECONOMIC JOURNAL

[DECEMBER

net income for an employed worker of type i is (i T,p) w^ + b^. It is then easy to show that (22') becomes ]a^(xq,[i-fi*{T,,r,,)] \ r + s + qfX \ / b, ^ b^ \ i\i-Tf I-TJ'

As the coefficient on bf/{i-Tf) +by,/{i - T ^ ) is negative, it follows by an argument similar to that in footnote i on p. 971 that an increase in bj or b^, from zero will improve the efficiency of the equilibrium. The relative effectiveness of these two instruments depends on the relationship between the tax rates - if Tf > Ty,, then bf is relatively more effective, and vice versa. In fact, for any positive firm (worker) employment subsidy there exists a critical tax rate, T]? {T%), such that for all Tf ^ T^, (T^ ^ TV) the equilibrium is efficient. The reason for this is simply that as long as bf is strictly positive, the right-hand side of (22") goes to minus infinity as Tf goes to i, and similarly for T^,. This has the following policy implication; it is possible, by an appropriate choice of profit tax and firm employment subsidy {bf > o), to run an unemployment benefit scheme (i^ < o) which has almost no negative effect on output or employment. The intuitive explanation for this is simply that by making the profit tax arbitrarily large relative to the wage tax, the incentive effects of a firm employment subsidy relative to unemployment benefit can be made arbitrarily large. Finally, it is worth comparing these results with those obtained by Pissarides (1985*) for a stochastic matching model with an endogenous unemployment vacancy ratio, but where workers and firms are homogeneous, and where vacancies are determined by a zero profit condition. In this setting, employer subsidies and unemployment benefits have very similar effects to the ones noted above; the former increases output and employment, the latter decreases it. The mechanism by which this occurs is, however, completely different. In the case of unemployment benefit, for example, an increase in unemployment benefit raises the wage, so the profitability of a vacancy falls below zero, so vacancy numbers fall also to restore equilibrium. This lowers the matching probabilities of workers, and so raises unemployment.
V. CONCLUDINO REMARKS

Two things have been accomplished in this paper. First, and most importantly, we have shown that search equilibrium may be inefficient even in the absence of external effects that work through the matching probabilities. By contrast, we have shown that the presence of high-skill types can lower the re-employment chances of low-skill types in equilibrium, even though this is not socially desirable. Secondly, we have considered policies that might counteract these external effects. Apart from policies that directly increase worker bargaining power, we have seen that a wage tax, or a profit tax combined with an employment subsidy improves efficiency, but that a profit tax alone may reduce it. It is worth remarking that, because of the symmetry of the model, an exactly similar kind of inefficiency will arise in the case where workers are homogeneous

1986]

' N A T U R A L ' RATE OF UNEMPLOYMENT

973

but firms differ in productivity - i.e. matches with low-productivity firms will not be accepted. Indeed, with heterogeneity on both sides of the market, one will still find non-acceptance of matches with low-productivity partners, so our results on the direction of the externality are fairly robust. Finally, an obvious limitation of the model is the assumption that worker abilities are observable by firms. If they are not, then there are two possibilities. Either there is a pooling equilibrium, where all ability types are hired and efficiency is restored, orfirmsmay attempt to screen workers. One way in which they may do this in practice is by using the duration of unemployment as a signal of productivity, and the incorporation of this kind of screening into the model is clearly a topic for future work. Birkbeck College Date of receipt of final typescript: February ig86
APPENDIX

Proof of the Proposition First, from (29), the a^ + Oi* are constant across i. This implies that a^-t-^f" are increasing in i. To see this, note that 6^ = a^- c, where c is some constant, so dropping the index i for convenience, ^^^_ef{d)dd\l[i-f{a-c)], so that ^ ^ ^ = i+/W/[i-^W][^'=-(-O] >o;

as the a,- are increasing in i, then so is a^ + Of". Furthermore, asfl^ f < oo by definition, a^ + O^'" is strictly increasing a&f{Oi) > o. Now, given this, it follows from (29) and (6) that there exists a A ; < i < n such that for for i ^ I, i^i. (A i) (A 2)

As a^ + 6i^ is strictly increasing, at least one of each of (A i) and (A 2) must hold strictly. Now, from (6), ^| is a solution to a^ + d = 4>i{6) As ^^ is continuous on the support of ^, and dip^/dd = 0 at any solution to a^ -1-^ = ^i{6), it follows that if a^ + d = ^i{6) has a solution, it is unique. Also, Hm9_a, <l>i{d) = o, and if di + O = (l>i{6) has a solution, ^^(o) ^ a^. (The latter follows from the fact that if <}>i{o) < a,., then a^-l-^ = <l)i{d) would have multiple solutions, an impossibility.) Therefore, 5 g ^ | as ^ + afg$if(^), which combined with (A i) and (A 2), completes the proof.

974

THE ECONOMIC JOURNAL

REFERENCES

[DECEMBER I986]

Bintnore, K. G. and Herrero, M. J. (1984). 'Frictionless non-Walrasian markets.' ICERD Discussion Paper No. 84/103, London School of Economics. Diamond, P. A. (1980). 'An alternative to steady-state comparisons.' Economic Letters, vol. 5, pp. 7-9. (igSi)- 'Mobility costs, frictional unemployment and efficiency.' Journal of Political Economy, vol. 89, pp. 798-812. ('982). 'Wage determination and efficiency in search equilibrium.' Review of Economic Studies, vol. 59, pp. 217-27. Friedman, M. (1968). 'The role of monetary policy.' American Economic Review, vol. 58, pp. 1-17. Johnson, G. and Layard, R. (1984). 'The natural rate of unemployment; explanation and policy.' Gentre for Labour Economics Discussion Paper, no. 225, London School of Economics. Pissarides, G. A. (1983).' The allocation ofjobs through search; some questions of efficiency.' Discussion Paper No. 156, Gentre for Labour Economics, London School of Economics. (1984). 'Efficient job rejection.' ECONOMIC JOURNAL, vol. 94, Gonference Supplement, pp. 97-107. (1985a). 'Short-run dynamics of unemployment, vacancies, and real wages.' American Economic Review, vol. 75, pp. 676-91. (1985*). 'Taxes, subsidies and equilibrium unemployment.' Review of Economic Studies, vol. 52, pp. 121-35. Rubenstein, A. and Wolinsky, A. (1984). 'Equilibrium in a market with sequential bargaining.' IGERD Discussion Paper 84/91, London School of Economics. Sargent, J.J. (1979). Macroeconomic Theory. Academic Press. Tobin, J. (1972). 'Inflation and unemployment.' American Economic Review, vol. 62, pp. 1-17.