Академический Документы
Профессиональный Документы
Культура Документы
George Symeonidis*
University of Essex and CEPR
Abstract. I examine the factors facilitating or hindering collusion using a comprehensive data
set on the incidence of price-fixing across UK manufacturing industries in the 1950s. The
econometric results suggest that collusion is more likely the higher the degree of capital
intensity and less likely in advertising-intensive than in low-advertising industries. There is
also some evidence of a non-monotonic relationship between market growth and the
likelihood of collusion. There is no clear link between concentration and the incidence of
collusion.
Keywords: Cartels, collusion, UK manufacturing.
JEL classification: L1.
*I would like to thank V. Bhaskar, Tim Hatton, Ken Hendricks (the editor), Stephen Martin,
Andrew Oswald, three referees, and seminar participants at Warwick, East Anglia, Essex, the
1997 conference on "Empirical Issues in Industrial Organization" in WZB, and the 1998
Royal Economic Society, EARIE and EEA conferences for helpful comments on earlier
versions.
Address for correspondence: Department of Economics, University of Essex, Wivenhoe Park,
Colchester CO4 3SQ, U.K. E-mail: symeonid@essex.ac.uk
I. Introduction.
Despite the substantial theoretical literature on the factors facilitating or hindering collusion,
empirical studies of the determinants of cartel formation and sustainability using direct
evidence, rather than relying on profitability indices to infer the possible operation of
collusive arrangements, are rare. Moreover, these studies have produced somewhat mixed
results. Thus while the findings by Hay and Kelley [1974] suggest that product differentiation
hinders collusion, this is not confirmed by Dick [1996a]. Collusion is more likely in
concentrated industries according to Hay and Kelley [1974] or Fraas and Greer [1977], but
not according to Dick [1996a] or Asch and Seneca [1976]. And while Dick [1996a] finds a
positive link between capital intensity and the incidence of collusion, Hay and Kelley [1974]
find no link between the ratio of fixed to total costs and collusive conduct. A possible
shortcoming of these studies is that the data either originate from antitrust cases, and may
therefore be subject to selection bias, or relate to export cartels, which cover only a very small
fraction of total economic activity. Also, some of the earlier studies rely on correlations rather
than regression analysis for their results, so they may not adequately control for links between
1
The present study examines the impact of several structural industry characteristics on
pricing conduct using an unusually comprehensive data set on the incidence of collusion
There is also an empirical literature on the related but somewhat different issue of cartel
duration (Jacquemin et al. 1981, Suslow 1991, Marquez 1994, Dick 1996b). These studies
have focused on the impact on cartel stability of various organisational features of cartels,
demand uncertainty, and the business cycle, in addition to structural industry characteristics
such as concentration. Probably the most robust prediction from this line of research is the
negative effect of demand uncertainty on cartel stability.
registered under the 1956 Restrictive Trade Practices Act, as well as several other sources on
competition in British industry in the 1950s, it is possible to identify those industries that
were subject to restrictive agreements at the time the Act was introduced. The final sample
used for the econometric analysis of collusion in this paper contains some 150 four-digit
industries, nearly half of which were subject to collusive pricing agreements in the mid- and
late 1950s.
The econometric results, based on a comparison of cartelised and non-cartelised
industries, suggest the following. Collusion is more likely the higher the degree of capital
intensity. It is less likely in advertising-intensive industries than in low-advertising industries.
There is some evidence that collusion is more likely under moderate market growth than in a
market with declining or stagnant demand, but less likely under fast growth than under
moderate growth. There is no clear link between concentration and the likelihood of collusion
once one controls for capital intensity, although an inverted-U relationship is present when no
account is taken of the potential endogeneity of concentration. Finally, there is weak evidence
that collusion may be less likely in R&D-intensive industries than in low-R&D industries.
The paper is structured as follows. Sections II and III describe the state of competition
in British manufacturing in the 1950s and the construction of the data set. Section IV
introduces the econometric model and section V presents the results. The final section
concludes.
An early study of collusive pricing in the UK in the 1950s by Phillips [1972] did not obtain
restrictions regarding capital investment, media advertising or R&D expenditure. Also, there
In fact, media advertising was not regulated in any advertising-intensive industry, even
though restrictions relating to sales promotion (e.g. gifts and coupons to consumers, financial
inducements to distributors, participation in exhibitions, advertising in trade catalogues, and
so on) were far more common. Expenditure on R&D was non-cooperatively determined in
all but one or two industries. There was some degree of technical cooperation, in the form of
exchange of technical information between firms, in some industries, and sometimes even
some joint R&D. Or patent-owning firms had to grant licenses to any other firm participating
in the arrangement. In all but one or two cases, however, these schemes do not seem to have
amounted to anything close to co-operative determination of R&D expenditure.
were no significant restrictions on entry in most cartelised industries. On the other hand,
some agreements contained restrictions of little significance in terms of their impact on
competition, such as "conditions of sale" (e.g. cash discount terms, delivery charges, nonprice matters, and so on), without any regulation of prices or trade discounts, even though
such regulation would have been legal under the circumstances. Finally, in some industries
the significance of the restrictions is uncertain, as in the case of agreements providing for the
central notification by the parties of inquiries received from customers and for some form of
uniform action, other than price-fixing, with respect to these inquiries.
Were the agreements effective? Effectiveness depended, first, on the extent to which
the parties conformed to the agreement or accepted the recommendations, and, second, on the
extent of competition from outside firms, domestic or foreign. On the whole, most
agreements seem to have been operated honourably and in most cases the parties accounted
for a large fraction of the relevant market. In particular, Swann et al. [1973, 1974], who
conducted case studies of 40 industries that had been subject to collusion in the 1950s, report
that all but one of these agreements had been operative before their cancellation. Swann et al.
also found that, in the large majority of cases, the parties accounted for at least 75% of the
relevant market. Heath [1963] states that 80% of the 159 agreements covered in his
questionnaire survey of the short-run impact of the 1956 Act were reported as being effective
at the time of cancellation, according to the replies received from firms and trade
associations, while, of the rest, only one in five had been ineffective for many years. On the
other hand, according to the Political and Economic Planning report on industrial trade
See Symeonidis [2002] for details on this point and Fershtman and Pakes [2000] for a recent
associations, carried out just before the 1956 Act was passed, some buyers could purchase at
prices lower than the agreed collusive prices, either from association firms or from outside
firms, in about half of the cartelised industries. It was also pointed out in the report, however,
that non-association firms were often unable to supply the product in the quantity or quality
required. More generally, the effectiveness of outside competition was limited in many
industries by a number of factors: the cartels tended to contain most or all of the largest and
best-known domestic firms; practices intended to limit outside competition such as
aggregated rebates and collective exclusive dealing were common; and competition from
imports was often limited because of tariffs and quantitative controls, differing technical
standards, transport costs or international restrictive agreements.
In conclusion, even though the "degree of collusion" must have varied across
cartelised industries depending on the type of restrictions, the extent of outside competition,
the balance of interests within the cartel, and so on and there is even some uncertainty about
the effectiveness of particular agreements, it is possible to split the UK manufacturing sector
in the 1950s into a group of cartelised industries and a group of industries without restrictive
agreements. Going beyond this for instance by classifying industries with respect to the
"degree of collusion" seems very difficult given the information available. Still, the binary
classification adopted here is sufficient for analysing the structural industry characteristics
that have facilitated or hindered collusion in the UK during the 1950s.
Collusion
The most comprehensive source of data on collusion in UK manufacturing in the mid- and
late 1950s is the Register of Restrictive Trading Agreements created under the 1956
Restrictive Trade Practices Act. The 1956 Act required the registration of restrictive
agreements between firms on goods, including both formal, written undertakings and
informal, verbal or even implied arrangements. Registered agreements were presumed to be
against the public interest and should therefore be abandoned, unless they were successfully
defended by the parties in the newly created Restrictive Practices Court as producing positive
benefits which outweighed the presumed detriment (or unless they were considered by the
Registrar of Restrictive Trading Agreements as not significantly affecting competition). For
those agreements that were brought before the Court and were defended by the parties, there
is also supplementary, and much more detailed, information in the reports of the Court.
Needless to say, agreements brought before the Court and defended by the parties can be
safely assumed to have been effective in the 1950s.
The reports of the Monopolies and Restrictive Practices Commission (MRPC) before
1956 and those of the Monopolies Commission (MC) after that date also contain detailed
information on the operation of cartels in particular industries. It is clear from these reports
that the industries investigated (some of which did not register any agreements under the 1956
Act) had been practicing collusion effectively during the 1950s. Other fully reliable sources
on the state of competition in the 1950s include the MRPC report on collective discrimination
(MRPC 1955), and several industry studies contained in Burn [1958]. Finally, two very
important but not fully reliable sources are the Board of Trade annual reports from 1949 to
1956 (Board of Trade 1949-1952, 1953-1956) and the Political and Economic Planning
[1957] survey of industrial trade associations, as well as unpublished background material for
this survey. These two sources provide information on industries alleged to be collusive; most
of these registered agreements, although some did not. This information must be treated with
caution because it is mainly based on complaints or reports given by buyers, and buyers may
wrongly deduce the existence of a price-fixing agreement from price uniformity or parallel
pricing. On the other hand, it may well be the case that some at least of the industries that
were alleged to be collusive and did not register any agreements were in fact collusive. Thus,
there is some uncertainty regarding the state of competition for products contained in the
Board of Trade annual reports or the P.E.P. survey but not mentioned as being the subject of
5
These data sources allow us to distinguish three groups of products: a group with
explicit restrictive agreements, a group without explicit restrictive agreements, and a group
with uncertain state of competition in the 1950s. The first group contains all products for
which there are registered agreements or which are reported as having been subject to
collusion in the MRPC and MC reports or any of the other fully reliable secondary sources.
The second group contains products not mentioned in any of the various sources. And the
third group contains products that do not appear in the Register of Restrictive Trading
Agreements or in any of the other fully reliable sources, but may have been subject to explicit
or tacit collusion according to the Board of Trade annual reports or the P.E.P. survey. This
5
Some of the information in the P.E.P. survey comes directly from firms that were
themselves parties to agreements. This information can be treated as reliable, and hence be
used to identify a few industries that were certainly collusive but failed to register their
agreements.
classification is based on certain assumptions as to the reliability of the various data sources
that seem quite reasonable. There is also, however, a more fundamental assumption implicit
in this classification: that the existence of a known (i.e. documented) explicit price-fixing
agreement is a good overall indicator of collusive conduct in British manufacturing industry
in the 1950s. One might raise a number of objections to this, and it is now necessary to
discuss these in some detail.
One possible objection is that the data on British cartels relate to explicit collusion,
but not to tacit collusion. Could it be the case that some of the industries classified as noncollusive actually practiced tacit collusion? Although one cannot rule out the possibility that
firms in an industry colluded tacitly in the absence of any explicit arrangement, it is difficult
to understand why this would have occurred given that explicit collusion was legal and
widespread in the 1950s. Furthermore, the 1956 Act required the registration of informal and
even "implied" understandings as well as formal agreements, and this seems to cover cases of
tacit collusion. Of course, it is still possible that colluding firms might have chosen not to
register but this is a different issue, to be discussed presently. As far as tacit collusion is
concerned, I would argue that, for all the above reasons and given that collusive arrangements
of all kinds were not enforceable in the courts, the distinction between tacit and explicit
collusion is not very important in the present context.
Another possible objection is that some of the agreements may have not been
effective at the time they were registered, so classifying these industries as collusive
introduces measurement error or even selection bias, if some of the industries that registered
explicit agreements may have done so precisely because collusion was difficult to sustain
otherwise. Are the above arguments valid? I think that the answer is in the negative, and I will
offer several different arguments to support this claim.
First, the case-study evidence discussed in the previous section strongly supports the
view that the large majority of the agreements had been effective. Second, the Register is not
the only source of information on collusion in British industry. Several industries were
investigated by the MRPC during the 1950s, and several more defended their agreements
before the Restrictive Practices Court. The available information leaves no doubt as to the
effectiveness of these agreements, which are a significant part of the total number. Third, all
the sources of information on the effects of the 1956 legislation emphasise that competition
was slow to emerge in many industries, and that this was often due to the fact that
information agreements replaced the former price-fixing arrangements in the short run. This
is not consistent with the view that these arrangements were not effective before 1956.
Fourth, a weak agreement could not expect to gain much from a favourable Court decision,
because it would still not be enforceable at law. So it is not at all clear why industries with
weak agreements would have a strong incentive to register. On the contrary, one might argue
that it was industries with strong agreements that had the strongest incentive to register and
try to maintain collusion, because of the potentially large cost of a cartel breakdown. Fifth,
the evidence from Lydall [1958] and Board of Trade [1946] discussed below also suggests
that there is no serious measurement error or selection bias in the construction of the collusive
group of industries in the present study.
The above discussion suggests that the assumption that the existence of an explicit
price-fixing agreement is a good overall indicator of collusive conduct is not an unreasonable
one in the present context. There is, however, one final difficulty, and this relates to the issue
of non-registration. In particular, non-registration of agreements, if widespread, would lead to
serious measurement error in the data and could even result in sample selection bias, if
certain types of industries had a stronger incentive to avoid registration than others.
More precisely, one can distinguish between two possible reasons for non-registration:
firms may simply suspend an agreement or they may switch to secret or tacit collusion. Take,
first, the former case. A reasonable conjecture in this case is that very weak agreements would
be more likely to be dropped immediately than stronger ones. Even if that were true, we
would not be losing much by failing to identify such cases, since we are interested in effective
agreements, not ineffective ones. But it is not even clear why the decision to immediately
cancel an agreement rather than register it should have occurred in certain types of industries
more than in others: there is not much to be lost by registering an agreement, even a weak
one, when the alternative is cancellation. In fact, many of the agreements that were not
registered were those that had been condemned by the MRPC. Clearly, these were not weak
agreements, but the parties must have thought that they had practically no chance of success
in the Court and wished to avoid further adverse publicity. This leaves us with the second
reason for non-registration mentioned above. One might argue, for instance, that industries
where tacit or secret collusion would be easier to sustain or less easily detected after 1956 had
less of an incentive to register. Failure to identify such cases could cause sample selection
bias.
An important thing to note with respect to the issue of non-registration is the historical
context of the introduction of the 1956 Act and, in particular, the uncertainty about the way
the legislation would be implemented. Because the attitude of the Court could not be known
until the first cases had been heard, firms were prompted to register their restrictive
agreements rather than drop or secretly continue them, although in some cases they redrafted
their agreements or even removed some of the restrictions in an attempt to increase the
likelihood of a favourable Court decision (see Swann et al. 1974, Hunter 1966). It seems
therefore that firms genuinely thought that they had a good chance of success in the Court,
10
which also explains why several agreements were defended in the Court despite the first few
6
The first restrictive agreements came before the Court in 1959 and were struck down.
Nearly all agreements were subsequently abandoned. Symeonidis [1998, 2002] discusses the
significance of the 1956 Act within the broader context of the evolution of UK cartel policy.
Symeonidis [2000a, 2000b, 2002] examines in detail the effects of the Act on firm conduct,
market structure and performance.
11
maintenance or collective exclusive dealing, without any fixing of common prices or market
shares. Many of these would normally not be registered, because the 1956 Act contained an
outright prohibition of the collective enforcement of resale price maintenance. Moreover,
collective exclusive dealing had been consistently condemned by the MRPC and was
therefore dropped by several industries prior to registration of their agreements. Once this is
taken into account, the incidence of non-registration of registrable agreements appears to have
been rather low.
Moreover, and most important, in this paper I do not rely solely on information about
registered agreements. As pointed out above, several of the sources examined allow us to
identify industries that either were certainly collusive or were alleged to be collusive, and did
not register any agreements. Although some of these sources are not perfect, it would really
be surprising if there were a significant number of cases that escaped all of them.
Finally, additional evidence suggesting that there should not be any significant sample
selection bias in the data from ineffective or unregistered agreements comes from two other
sources. The first is a questionnaire survey of competition in UK manufacturing in the 1950s
(Lydall 1958). This study, which used a sample of 876 manufacturing firms from all sectors,
did not specifically examine collusion; however, some of the information provided suggests
that firms that perceived their condition as being characterised by "no strong competition"
were primarily in industries which had a high incidence of explicit collusion, according to my
classification, while firms that thought that they were facing "strong competition" were
chiefly in industries without many agreements. The second source is a survey of UK cartels
carried out in the mid-1940s by the Board of Trade (Board of Trade 1946). Although the
survey was not fully comprehensive, the industries chosen spanned the whole spectrum of
manufacturing industries, covering capital-good as well as consumer-good industries. Despite
12
this large coverage, and the fact that most of the British cartels of the 1950s had already been
active at the time of the Board of Trade survey (see Swann et al. 1974), there are virtually no
industries reported as being subject to restrictive agreements in the early 1940s which are not
mentioned as collusive in at least one of my data sources for collusion in the 1950s.
Furthermore, nearly all of these industries registered agreements under the 1956 Act.
It seems reasonable to conclude, on the basis of the above discussion, that the issue of
potential non-registration of agreements does not cause any significant bias or measurement
error in the present data. Admittedly, it is not possible to back up this claim with anything
more than the indirect evidence discussed above. In this paper therefore I will also examine
the robustness of the econometric results to an alternative classification of the industries
according to their competitive status. In particular, I will compare the benchmark
classification set out above with an alternative one designed to maximize the chances of
identifying any selection bias in the data. For this alternative classification, there will be again
three groups of industries. The collusive group will now contain all products which are
reported as being subject to collusion in any one of the following four key sources: the reports
of the MRPC and the MC, the reports of the Restrictive Practices Court, the Board of Trade
annual reports, and the P.E.P. survey. The competitive group will contain products not
mentioned in any of the various sources. And the group with uncertain state of competition
will contain products for which there are registered agreements but these are not mentioned in
any of the four key sources mentioned above. Note that registered agreements not mentioned
in any other source are probably those with a higher chance of not having been effective.
Also, industries without registered agreements but mentioned in the Board of Trade annual
reports or the P.E.P. survey as collusive are precisely those where collusion is more likely to
have occurred even though no agreements were registered. Thus a comparison of the two sets
13
of results is a very powerful indirect check for selection bias caused by unregistered or
ineffective collusive agreements.
In addition to an assessment of the reliability of the various data sources, I used two
other criteria to classify industries as "collusive", "competitive" or "ambiguous": (i) the types
of restrictions; and (ii) the proportion of an industry's total sales covered by products subject
to agreements and, for each product, the fraction of the UK market covered by cartel firms. In
particular, I classified the various types of restrictions as significant, non-significant or
7
uncertain, according to their likely impact on competition. Next, I assigned the products that
were subject to agreements to the various headings of the classification used in the 1958
concentration statistics. It was sometimes the case that certain products within a particular
four-digit industry were subject to significant restrictions, while others were not. An industry
was classified as collusive if the products subject to significant restrictions accounted for
more than 50% of total industry sales revenue. It was classified as competitive if the products
subject to significant or uncertain restrictions accounted for less than 10% of industry sales
revenue. And it was classified as ambiguous in all remaining cases. In fact, most industries
classified as competitive were free from any restrictive agreements. I used the 10% cut-off
point because in some cases secondary industry products were subject to restrictive
agreements, although core industry products were not. Similarly, most industries classified as
collusive had agreements covering all industry products. I used the 50% cut-off point because
in some cases most core industry products were subject to price-fixing, though some were
not. However, I also tried alternative cut-off points to check the robustness of the results. I
this classification.
14
assumed, in the case of agreements of nationwide application, that the parties accounted for a
substantial fraction of the relevant market. There were also a few cases of important regional
agreements, and an estimate of the fraction of industry sales subject to restrictions was then
made.
All industries classified as ambiguous were excluded from the sample. Collusion
could then be modelled by means of a dummy variable taking the value 1 for collusive and
the value 0 for competitive industries. Following my benchmark classification of industries, I
constructed the variable COLL1; in this case the sample after also excluding industries for
which concentration ratios were not available or not relevant (see below) contained 151
four-digit industries, of which 71 were collusive and 80 were competitive. Following my
alternative classification of industries, I constructed COLL2; in this case the sample contained
140 four-digit industries, of which 60 were collusive and 80 were competitive. While the
group of competitive industries was the same for COLL2 and COLL1, only 51 industries were
in the collusive group under both classifications.
Table I summarises the available information on the incidence of collusion across
manufacturing sectors in the 1950s according to the classification criteria used for
constructing COLL1. (Applying the criteria for COLL2 does not significantly alter this
picture.) The industry definitions used in this table are taken from the official concentration
statistics for 1958 and correspond to Census of Production four-digit industries. The 212
industries included in the table cover most of the manufacturing sector, although certain
sectors, such as chemicals or instruments, may be somewhat under-represented, and others,
such as basic metals or textiles, may be slightly over-represented.
Since this sample will also be used to examine the links between the incidence of collusion
15
As can be seen from this table, collusive agreements were much more widespread in
some sectors than in others. In particular, sectors with a high incidence of collusion
(measured as the fraction of collusive industries in the total number of industries in any given
sector) include mining and quarrying, basic metals, other metal products, building materials,
and electrical engineering. Sectors with a low incidence of collusion include leather goods,
clothing and footwear, and instruments. In the case of chemicals, the relatively low incidence
of collusion shown in Table I is to some extent a feature of the particular sample used, since a
number of chemical industries with restrictive agreements are not included in the sample.
However, the figures also reflect the absence of collusive agreements in most of the
advertising-intensive, consumer-good industries in the chemical sector. Of the other sectors,
paper, printing and publishing has an above-average incidence of collusion; textiles and
mechanical engineering are intermediate; and food and drink, shipbuilding and vehicles,
wood products, and other manufacturing have a below-average incidence of collusion.
and other variables, I excluded industries without available concentration data for 1958.
However, there is no reason to believe that this causes any significant bias with respect to
comparisons across sectors. In addition, I excluded four industries with significant
government participation or intervention in the 1950s (sugar, ordnance, aircraft, locomotives),
two industries that hardly existed prior to the mid-1950s (synthetic rubber, tufted carpets),
and one where imports were very much higher than domestic production and were subject to
regulation (iron ore).
16
Other variables
The concentration measure used, CONC5, is the share of the five largest producers in the total
sales revenue of UK firms. I used Census-based data at the four-digit (or "product group")
level of aggregation for 1958, a year when the agreements were still in place. This level of
9
aggregation seems appropriate for the analysis of the determinants of collusive pricing.
However, when competition tends to operate at the regional rather than the national level,
these concentration ratios may be a poor measure of market structure. I made an effort to
identify such cases; five industries were thus excluded from the samples used.
A measure of market growth at the four-digit industry level between 1954 and 1958 is
the ratio of 1958 industry sales revenue at net producer prices to 1954 industry sales revenue,
deflated using 16 sector-specific producer price indices. This measure, GROWTH, goes some
way into controlling for changes in relative prices, in particular those induced by changes in
the relative cost of materials (a significant factor during the 1950s). An alternative measure of
market growth, constructed by deflating the sales figures by the general producer price index,
was also tried, as were measures of market growth between 1951 and 1958. All these
produced results similar to those reported in section V below. Note that Census of Production
data in the 1950s at the level of aggregation used in this paper are only available for 1951,
1954 and 1958. The 1954-1958 period is a relevant one as most agreements were registered in
1957, and it is long enough to ensure that the change in sales reflects the trend in demand
9
In some cases, a four-digit industry may comprise two or more sub-markets which are
largely independent on the demand side. I identified 8 such cases in my samples. Three were
in the "collusive" group and five in the "competitive" group, so it is unlikely that any bias is
caused from this factor.
17
10
Two measures of capital intensity in 1958 were defined. The first of these, CAPINT1,
is the total value of capital stock in the industry divided by the number of plants, i.e. the
capital stock of the average plant. The second, CAPINT2, is the capital-labour ratio. The data
on capital stock are estimates rather than primary data and were taken from O'Mahoney and
Oulton [1990]. They were only available at the three-digit level of aggregation (i.e. for
Census "industries"), but it was often possible to derive reasonable approximations of capital
stock at the four-digit level by using Census of Production data on the fraction of the threedigit industry investment accounted for by each four-digit industry within any given threedigit industry.
The extent of foreign competition may have also been a factor influencing the
likelihood of collusion between UK firms in any particular industry. Constructing an adequate
measure of foreign competition is very difficult, however (see the Appendix for more on this
point). Here I use a somewhat crude measure of the extent of (potential or actual) foreign
competition, constructed on the basis of the Kitchin [1976] estimates of effective protection
rates for 1963 and other available information for the 1950s. This is a dummy variable,
FOREIGN, taking the value 0 for industries with higher protection in the mid-1950s and the
value 1 for industries with lower protection.
10
11
One reason for using 1954-1958 growth rather than 1951-1958 growth in the benchmark
specification is that price and output controls were still effective in 1951 in a number of
industries, and this may have distorted growth rates during 1951-1954. Any such remaining
controls had generally been withdrawn by 1954.
11
The former group contains the engineering industries, instruments, vehicles, finished metal
goods, some chemicals, paper and paper products, furniture, pottery and glass, most finished
18
factors that may have influenced the extent of foreign competition across industries such as
transport costs or the scope for restrictive practices aimed at reducing outside competition.
Because of possible reservations regarding the use of such a crude proxy, I have run
regressions both with and without it. As it turned out, the coefficient on FOREIGN always
had the expected sign and the results did not change significantly when this variable was not
included among the regressors.
Finally, advertising-sales ratios and R&D-sales ratios, at the four-digit level, were
used to classify the industries in the sample. The industries were split, on the one hand, into
low-advertising and advertising-intensive using the 1% advertising-sales ratio as a cut-off
point, and, on the other, into low-R&D and R&D-intensive using the 1% R&D-sales ratio as a
cut-off point. The 1% cut-off point was chosen since it is commonly used to classify
industries according to advertising or R&D intensity. In addition, it resulted in a not-toounbalanced split, whereas a higher cut-off point (2%, say) would leave only a small number
of industries in the advertising-intensive and R&D-intensive categories. Thus, of the 151
industries in the sample used in regressions with COLL1, 111 had an advertising-sales ratio
lower than 1% and 119 had an R&D-sales ratio lower than 1%. In any case, I also
experimented with using both the 1% and the 2% cut-off points to classify the industries
according to advertising intensity and R&D intensity, and these results will be discussed
below.
There are two main reasons why the advertising-sales and R&D-sales ratios were used
textile goods, rubber products, and various other finished manufactures. The latter group
contains most food and drink industries, some chemicals, basic metals, clothing and footwear,
wood products, publishing, leather and most textile semi-manufactures, and building
materials.
19
to classify the industries rather than being used themselves as continuous explanatory
variables in the regressions. On the one hand, data limitations made it difficult to construct
very precise figures, especially for advertising intensity of low-advertising industries and for
R&D intensity. On the other hand, these binary classifications reflect exogenous industry
characteristics, such as advertising effectiveness or technological opportunity, but are
relatively robust to other factors that may affect advertising or R&D intensity. Hence the
dummy variables resulting from these classifications can be treated as exogenous variables, a
point that will be further elaborated in the next section.
The empirical model I will use is a probit model with the basic specification
12
The impact of some other factors, such as demand uncertainty, is difficult to evaluate due
to data limitations. Also, since the restrictive agreements were in most cases long-standing
and systematic cross-industry information on breakdowns of collusion is not available, the
impact of business cycle fluctuations on cartel stability cannot be assessed in the present
context. Note, however, that some of the factors mentioned in the theoretical literature as
hindering cartel stability, such as the presence of large buyers or the lumpiness of orders, are
probably not very relevant in the present context because public bodies, which were the
largest buyers, did not do much to deter collusion in the 1950s.
20
5 ln CAPINT 1i 6 CONC5i 7
CONC5 i
8 FOREIGN i ui
A possible objection to the use of GROWTH in the above specification is that this
variable may be endogenous. In particular, a two-way link between growth and collusion
cannot, in principle, be ruled out. Unfortunately, there is no appropriate instrument for
GROWTH: as COLL1 essentially does not vary over the 1950s, lagged values of sales growth
cannot be used as instruments. However, there is some indirect evidence supporting the
hypothesis that growth during 1954-1958 was largely determined by exogenous shifts in
demand rather than by the state of competition. The 1954-58 growth figures can be compared
(for industries with available data for both periods) to the corresponding figures for 1963-
21
1968, a period when the majority of the previously cartelised industries had already
abandoned their restrictive agreements. The median value of GROWTH was 1.057 for 64
industries which were cartelised in the 1950s and had switched to competition by 1968, and
1.069 for 78 non-cartelised industries. If collusion had had a significant (presumably
negative) effect on sales growth during 1954-58, the abolition of restrictive agreements
should affect (presumably reverse) this initial small difference in growth between previously
cartelised and non-cartelised industries. However, this is not what actually happened: the
median value of sales growth during 1963-68 was 1.148 for the 64 previously cartelised
industries, and 1.162 for the 78 non-cartelised ones. Thus it seems unlikely that collusion had
a significant effect on growth during 1954-1958 and there is no reason to believe that the
potential endogeneity of GROWTH is a serious problem in the present context.
Since ADV1 and RD1 essentially reflect exogenous industry characteristics, namely
advertising effectiveness and technological opportunity, respectively, they can also be treated
as exogenous variables. The idea here is that while the actual level of the advertising-sales
ratio and the R&D-sales ratio is endogenous and may depend on many variables, including
the intensity of price competition, it is generally exogenous industry characteristics that will
determine whether this ratio is above or below 1% or 2%. Thus, for an industry below the 1%
cut-off point, advertising/R&D is not an important strategic variable: in such an industry,
advertising is not very effective in raising consumers willingness to pay or there is little
scope for technological innovation from within the industry. On the other hand, in an industry
above the 1% cut-off point (and even more in an industry above the 2% cut-off point),
advertising/R&D "works". Of course, whether such an industry has an advertising-sales ratio
or R&D-sales ratio of 5% or 10% may be largely determined endogenously. But my binary
variables ADV1 and RD1 are not very sensitive to endogenous factors that affect advertising
intensity and R&D intensity. The assumption of exogeneity of ADV1 and RD1 is also
consistent with the fact that a comparison of advertising-sales ratios and R&D-sales ratios
across various years revealed very few instances where an industry had moved from below
1% to above 1% or vice versa; and in most cases this was due to an exogenous institutional
change, namely the introduction of TV advertising in the UK in the mid-1950s.
13
GROWTH, GROWTH , ADV1, RD1, lnCAPINT, FOREIGN and lnSALES, where SALES is
1958 four-digit industry sales revenue a variable which is typically used in studies of the
determinants of concentration as a proxy for market size, and is not endogenous with respect
to collusion in a cross section of industries since it is mainly determined by the industry
14
categories used. As it turns out, the two sets of results are similar, except for the coefficients
13
I also tried a classification using the 2% cut-off point, and obtained similar results. This
suggests that my results are not driven by industries with intermediate advertising or R&D
intensity. They are rather driven by industries with very low or very high advertising/R&D
intensity, and it is clear that very low or very high advertising/R&D intensity is a largely
exogenous feature of industries.
14
Recent studies of the determinants of concentration suggest that either separate regressions
should be run for exogenous sunk cost industries and advertising-intensive or R&D-intensive
and the t-statistics on CONC5 and CONC5 , which differ considerably between the two
models.
15 16
The first-stage results from the two-stage probit are presented in Table V. To
Vuong [1988] gave results very similar to those presented in table IV. This procedure also
allows for a simple Wald test for exogeneity. The test sometimes rejects the null hypothesis
of exogeneity, and sometimes it fails to reject.
16
ratio was 0.57 and the R in the regression of UK concentration on US concentration and a
constant was 0.33. Despite the small size of the sample, the results when using US
concentration as an instrument for UK concentration were similar to those reported in Table
IV (although the standard errors were generally higher), and the concentration variable was
nowhere significant whether the quadratic term was included or not. Of course, this latter
result may simply indicate that US concentration is a poor instrument for UK concentration.
Finally, note that it is not possible to use lagged UK concentration as an instrument in the
present case. Comparable concentration data for earlier years are not available; even if such
data were available, they would not constitute valid instruments, because the collusion
variable varies very little within any industry over the 1950s.
better interpret these results, note that the mean of lnSALES for the benchmark sample of 151
industries is 10.34, the standard deviation 0.89, the minimum 8.36 and the maximum 12.93.
As can be clearly seen from Table V, lnSALES is an important determinant of CONC5 in
these regressions.
[Place Tables III, IV and V approximately here]
serving
as
the
benchmark
category):
instruments
and
electrical
17
metal products; and building materials, pottery, glass and wood products.
GROWTH and GROWTH are, respectively, positive and negative, and they are both typically
statistically significant at the 5% or the 10% level (a joint test of significance gives similar
results). The value of GROWTH that corresponds to the maximum probability of collusion
typically ranges around 1.10-1.30. About 25% of the industries in the sample have growth
rates between 1954 and 1958 higher than this. Thus it would seem that while a moderate
growth rate is more conducive to stable collusion than a stagnant or declining demand, fast
growth hinders collusion. Note, however, that the effect of demand growth on collusion
appears much weaker or non-existent in regressions using COLL2. This may suggest that this
particular result may be driven by a relatively small number of industries and is therefore
17
In the two-stage model the reported standard errors are uncorrected, i.e. the fitted values
from the first stage are essentially treated as observed values of CONC5 for purposes of
inference. A formula for correcting the covariance matrix in two-stage probit models is given
in Maddala [1983, p. 245], but it cannot be used in regressions including the square of the
fitted value from the first stage (which is the appropriate specification in the present case).
However, the corrected standard errors that I computed using the Maddala formula in
regressions not including the square of the fitted value were very similar to the uncorrected
standard errors. This is consistent with Monte Carlo results comparing corrected and
uncorrected standard errors in two-stage probit models (Bollen et al. 1995).
18
How should one interpret the observed inverted-U relationship between demand
growth and the likelihood of collusion? A standard result of the game-theoretic literature is
that collusion is easier to sustain the higher the rate of demand growth because the larger is
then the importance of future profits from collusion relative to the current gain from defecting
(Tirole 1988). This mechanism may, however, be offset by other mechanisms under
conditions of fast growth. In particular, fast growth may be associated with higher demand
uncertainty and therefore destabilise collusion. Alternatively, fast growth may lead to
significant new entry and hinder the attempts of firms to coordinate on a collusive price or set
of prices. Finally, firms may be subject to capacity constraints which become binding when
demand grows fast, and this may hinder collusion by limiting the scope for punishing
defectors. Each of these mechanisms would result in a non-monotonic relationship between
demand growth and the likelihood of collusion: positive for values of growth below a certain
level, but negative for high levels of growth.
A possible objection to the above interpretation could be that new industries usually
grow rapidly, so that the observed negative link between fast growth and collusion might be
picking up the potential effect of industry age on collusion with new industries being less
likely to have cartels simply because firms need time to learn to coordinate with their
competitors. To check this hypothesis I excluded from the sample all six industries which
were established in the UK after 1940 and were therefore still relatively new in the late 1950s.
18
If the quadratic term GROWTH is dropped from regressions using COLL2, then the linear
term GROWTH has a negative coefficient, which, however, is usually not statistically
significant. In regressions using COLL1 the coefficient on GROWTH is never statistically
significant if the quadratic term is omitted.
without the quadratic term.) A possible reason for the absence of any clear link between
concentration and the incidence of collusion in the present context is that the coordination
and monitoring of collusion must have been greatly facilitated by the fact that the agreements
were not illegal and were often operated by members of trade associations. On the other hand,
a non-linear relationship could result from the presence of capacity constraints (see Brock and
Scheinkman 1985), or it could be due to the fact that high concentration is sometimes
associated with significant firm asymmetries or the presence of dominant firms, and these
factors may hinder collusion.
Could the results for concentration be affected by measurement error? The issue here
is that the data used to construct CONC5 are for domestic firms rather than for the domestic
market. This is not necessarily a problem in the present context since most restrictive
agreements were operated between UK firms rather than between firms selling in the UK
market. But to check whether the results are affected by this factor, I ran regressions
excluding from the sample all industries for which the value of imports as a fraction of sales
revenue by domestic firms was higher than 10% (this reduces the sample by about 15%). The
2
coefficients and t-statistics on CONC5 and CONC5 were similar to those reported here.
Finally, note that if the capital intensity variable is dropped, then the coefficient on
CONC5 is positive and highly significant. This suggests that capital intensity, rather than
concentration per se, could be the key factor in explaining the alleged high incidence of
collusion in highly concentrated industries, even when collusion is illegal (and thus
coordination and monitoring are more difficult). Could it be the case that the reverse is true,
i.e. that capital intensity is not really important for collusion and is essentially picking up the
effect of concentration in the present sample, since the two variables are correlated? I think
the answer is no. First, note that if there were a multicollinearity problem, then the standard
errors on capital intensity should be large, but this is not the case here. Second, recall from
my previous discussion that the result on capital intensity is robust to more or less
measurement error in that variable.
The results also suggest that the incidence of price collusion is lower in advertisingintensive industries: the coefficient on ADV1 is negative and everywhere statistically
significant at the 1% or the 5% level. Classification of an industry as advertising-intensive
reduces the probability of collusion (at the sample means) by 25-35%. Furthermore, the
observed negative association between advertising and collusion cannot be due to any
negative effect of collusion on advertising. In Symeonidis [2000b] I show that the
intensification of price competition following the abolition of cartels in the UK caused
advertising intensity to fall. This is consistent with the idea that price collusion may lead to
more advertising as firms intensify competition in other dimensions given that they do not
compete on price (see Scherer and Ross 1990). Thus the negative link between advertising
and collusion that I find in the present paper can only mean that collusion is less likely to
19
In separate regressions for producer-good and consumer-good industries, ADV1 still has a
20
collusion should not be much more difficult to sustain than in a low-R&D industry.
This
contrasts with the case of advertising-intensive industries, where heavy advertisers often coexist with firms that do not advertise but employ different sales strategies, so that large
differences in brand image are common (and presumably hinder collusion in price).
Some case-study evidence on R&D-intensive industries with price-fixing agreements
seems to be consistent with this latter interpretation. In some of these industries, such as
heavy electrical machinery or telecommunications equipment, the collusive agreements were
mostly operated by a small number of R&D-intensive firms, which were the only UK
producers of the products in question (see Swann et al. 1973, MRPC 1957, Hart et al. 1973).
Hence price collusion was presumably not hindered by large quality differences. Moreover, in
several other industries the agreements involved "patent pooling" and exchange of technical
information (but not cooperation in R&D). These schemes may have been used to limit
quality differences and ensure the success of the pricing arrangements.
Several checks of the robustness of the results to alternative ways of constructing the
dependent variable and to alternative proxies for some of the independent variables were
performed. First, I experimented with different cut-off points for the classification of
industries as collusive, competitive or ambiguous according to the fraction of total sales
covered by products subject to agreements. This did not change the sample very much,
anyway; for instance, changing the cut-off points from 10% and 50% to 20% and 80%,
respectively, results in 64 industries being assigned to the collusive group and 86 being
20
Differences in product quality between firms will also be small in industries with process
rather product R&D. This may also weaken any link between R&D intensity and collusion,
although several industries where process R&D is important, such as basic chemicals, are not
included in the sample used here.
assigned to the competitive group. For all these alternative definitions of COLL, the results
were similar to those reported here. Second, I included in the sample the ambiguous
industries by treating them as intermediate and performed ordered probit. This is not very
appropriate, given the heterogeneity of this group of industries, but it may be useful merely as
a robustness check. Again the results were similar to those reported in Tables III and IV (see
the Supplemental Materials in the JIE website for details).
Finally, I performed regressions using alternative proxies for advertising effectiveness
and technological opportunity. In particular, I used ADV12 and RD12 instead of ADV1 and
RD1, where ADV12 and RD12 take the values 0, 1 or 2 according to whether an industrys
advertising intensity and R&D intensity, respectively, are lower than 1%, between 1% and
2%, or higher than 2%. All the results were very similar to those reported in Tables III and IV
(see the Supplemental Materials in the JIE website).
The results of the present paper have significant implications for the various
theoretical approaches to collusion. First, the observed negative link between advertising and
collusive pricing is not consistent with theoretical results from models of cartel stability that
emphasise horizontal product differentiation. To interpret this result it seems necessary to
shift to models of cartel stability emphasising vertical differentiation or to focus more on the
issue of coordination rather than enforcement of collusion. Second, the apparent absence of
any clear link between concentration and the likelihood of collusion may not be surprising in
a context where coordination and monitoring were facilitated by the fact that the agreements
were not illegal. Even in this case, however, the standard game-theoretic prediction that high
concentration facilitates cartel stability by reducing the incentive to defect would be valid; but
it does not seem to get much support in the present sample. Third, capital intensity seems to
be an important determinant of the likelihood of collusive pricing, at least for manufacturing
industries, although its role has been little emphasised in the theoretical literature. More
specifically, the link between capital intensity and collusive pricing may suggest that more
attention should be given to the conditions of entry as a key factor in the formation and
stability of cartels.
Ambiguous
Competitive
14
12
Mechanical engineering
10
Instrument engineering
Electrical engineering
12
Textiles
15
76
56
80
Total
Collusive
(n = 71)
Non-collusive
(n = 80)
All
(n = 151)
GROWTH
Mean
St. deviation
Min.
Max.
1.08
0.20
0.71
1.59
1.17
0.45
0.59
3.24
1.13
0.36
0.59
3.24
lnCAPINT1
Mean
St. deviation
Min.
Max.
0.01
1.15
-2.98
2.17
-1.09
1.33
-3.60
3.42
-0.57
1.36
-3.60
3.42
lnCAPINT2
Mean
St. deviation
Min.
Max.
1.23
0.73
-0.99
3.10
0.64
0.86
-1.18
2.38
0.92
0.85
-1.18
3.10
CONC5
Mean
St. deviation
Min.
Max.
0.61
0.20
0.19
0.98
0.52
0.28
0.08
0.99
0.56
0.25
0.08
0.99
62
49
111
31
40
59
60
119
12
20
32
45
34
79
26
46
72
Table III. Regression results for the determinants of collusion: Probit estimation.
3.16
(1.32)
4.93
(1.75)
4.37
(2.11)
5.81
(2.71)
0.52
(0.28)
2.06
(1.02)
0.88
(0.44)
2.80
(1.34)
-1.47
(-1.63)
-2.13
(-1.88)
-1.77
(-2.56)
-2.31
(-3.13)
-0.45
(-0.69)
-0.97
(-1.28)
-0.66
(-0.99)
-1.28
(-1.81)
ADV1
-0.90
(-3.07)
-0.87
(-3.01)
-0.95
(-2.74)
-0.78
(-2.27)
-0.88
(-2.66)
-0.93
(-2.96)
-0.81
(-2.03)
-0.78
(-2.08)
RD1
-0.71
(-2.12)
-0.44
(-1.37)
-0.77
(-1.55)
-0.39
(-0.78)
-0.75
(-2.07)
-0.47
(-1.38)
-1.08
(-1.86)
-0.62
(-1.14)
lnCAPINT1
0.55
(4.32)
0.69
(3.94)
0.51
(4.00)
0.72
(4.42)
lnCAPINT2
0.54
(3.26)
0.73
(3.25)
0.45
(2.74)
0.82
(3.52)
8.53
(3.06)
8.52
(3.26)
8.97
(2.97)
8.68
(2.91)
5.82
(2.10)
5.88
(2.20)
6.02
(2.02)
5.66
(2.02)
-7.74
(-3.13)
-7.08
(-2.91)
-8.13
(-3.21)
-7.25
(-2.92)
-5.10
(-2.08)
-4.45
(-1.90)
-5.60
(-2.19)
-4.51
(-1.88)
FOREIGN
-0.54
(-2.10)
-0.62
(-2.41)
-0.60
(-1.36)
-0.63
(-1.53)
-0.65
(-2.45)
-0.69
(-2.65)
-0.40
(-0.90)
-0.41
(-0.95)
constant
-2.63
(-1.63)
-4.84
(-2.80)
-4.26
(-2.34)
-6.61
(-3.87)
-0.56
(-0.39)
-2.61
(-1.82)
-1.07
(-0.63)
-3.87
(-2.35)
sector dummies
No
No
Yes
Yes
No
No
Yes
Yes
% correct
predictions
78.8
74.8
80.8
79.5
76.8
71.5
78.1
74.2
1 - lnL / lnL0
0.338
0.290
0.438
0.389
0.311
0.263
0.422
0.373
No. of
observations
151
151
151
151
140
140
140
140
GROWTH
GROWTH
CONC5
CONC5
37
Table IV. Regression results for the determinants of collusion: Two-stage probit.
4.22
(1.70)
6.13
(2.54)
4.83
(2.27)
6.75
(3.07)
1.04
(0.51)
2.67
(1.36)
1.16
(0.58)
3.55
(1.64)
-1.80
(-1.85)
-2.32
(-2.52)
-1.92
(-2.70)
-2.50
(-3.48)
-0.60
(-0.81)
-1.00
(-1.45)
-0.77
(-1.12)
-1.43
(-2.01)
ADV1
-0.90
(-3.02)
-0.77
(-2.48)
-0.98
(-2.84)
-0.74
(-2.18)
-0.85
(-2.44)
-0.75
(-2.14)
-0.80
(-1.95)
-0.64
(-1.61)
RD1
-0.66
(-1.97)
-0.17
(-0.53)
-0.76
(-1.53)
-0.15
(-0.28)
-0.69
(-1.89)
-0.22
(-0.61)
-0.98
(-1.75)
-0.41
(-0.74)
lnCAPINT1
0.60
(3.50)
0.79
(3.29)
0.66
(3.58)
0.93
(3.60)
lnCAPINT2
0.89
(3.53)
1.18
(3.24)
0.88
(3.49)
1.40
(3.59)
2.71
(0.99)
1.40
(0.48)
1.47
(0.45)
-0.10
(-0.03)
2.21
(0.79)
5.37
(1.46)
2.27
(0.59)
2.54
(0.56)
-3.23
(-1.42)
-2.93
(-1.15)
-2.66
(-1.00)
-2.40
(-0.84)
-3.15
(-1.38)
-6.63
(-2.09)
-4.03
(-1.29)
-5.05
(-1.35)
FOREIGN
-0.57
(-2.22)
-0.70
(-2.58)
-0.49
(-1.23)
-0.48
(-1.23)
-0.64
(-2.37)
-0.67
(-2.44)
-0.36
(-0.80)
-0.26
(-0.58)
constant
-1.84
(-1.03)
-3.90
(-2.42)
-2.38
(-1.20)
-4.61
(-2.35)
0.33
(0.20)
-2.85
(-1.92)
0.39
(0.20)
-3.00
(-1.46)
sector dummies
No
No
Yes
Yes
No
No
Yes
Yes
% correct
predictions
80.8
72.8
78.8
76.8
79.5
72.8
78.1
74.2
1 - lnL / lnL0
0.300
0.259
0.407
0.372
0.302
0.267
0.421
0.386
No. of
observations
151
151
151
151
140
140
140
140
GROWTH
GROWTH
CONC5
CONC5
38
GROWTH
0.37
(2.56)
0.60
(3.41)
0.33
(2.14)
0.46
(2.62)
-0.10
(-2.34)
-0.15
(-2.94)
-0.09
(-2.12)
-0.13
(-2.49)
ADV1
0.03
(1.13)
0.04
(1.29)
0.01
(0.36)
0.04
(1.15)
RD1
0.03
(0.99)
0.11
(2.80)
0.08
(1.66)
0.13
(2.50)
lnCAPINT1
0.14
(14.61)
0.13
(11.43)
lnCAPINT2
0.18
(10.76)
0.16
(8.32)
lnSALES
-0.14
(-9.71)
-0.12
(-6.87)
-0.13
(-8.31)
-0.10
(-6.12)
FOREIGN
0.04
(1.52)
-0.002
(-0.07)
0.06
(1.68)
0.04
(0.88)
constant
1.81
(9.82)
1.11
(5.56)
1.70
(9.02)
1.12
(5.47)
No
No
Yes
Yes
0.68
0.56
0.71
0.63
GROWTH
sector dummies
R
39
APPENDIX
The concentration data for 1958 were obtained from Summary Table 5 of the 1963 Census of
Production. Data on manufacturers' sales revenue at current net producer prices were obtained
from the above table and from the individual industry reports of the Census of Production
(various years). The figures are sales by all firms employing 25 or more persons. Producer
price indices were obtained from the Annual Abstract of Statistics and the Historical Record
of the Census of Production 1907 to 1970. In some cases where information was not
available, approximate price indices were computed on the basis of indices of industrial
production published in the Annual Abstract of Statistics.
Capital stock was defined as plant and machinery. I used the net capital stock
estimates of O'Mahoney and Oulton [1990] for 1958. Although these are at the three-digit
level, it was often possible to compute estimates of four-digit capital stock by multiplying the
1958 three-digit capital stock by the ratio of four-digit investment to three-digit investment,
averaged over 1954 and 1958. Investment data as well data on the number of plants and
employment at the four-digit level of aggregation were taken from the individual reports of
the 1958 Census of Production. Some adjustments were made to ensure comparability in light
of the fact that the O'Mahoney and Oulton figures are based on the 1968 S.I.C., which is
somewhat different from the 1958 S.I.C. Firms employing less than 25 persons were not
taken into account.
R&D expenditure data are available for the early 1960s at a level of aggregation
between the two-digit and the three-digit and also for the mid- and late 1950s (although some
of the latter data are not very reliable). They have been published in Research and
Development Expenditure, Studies in Official Statistics no. 21 (HMSO, 1973); Industrial
40
41
42
43
REFERENCES
Asch, P. and Seneca, J. J., 1976, Is Collusion Profitable?, Review of Economics and
Statistics, 58, pp. 1-12.
Board of Trade, 1946, Survey of Internal Cartels (Board of Trade, London).
Board of Trade, 1949-1952, Monopolies and Restrictive Practices (Inquiry and Control Act,
1948: Annual report by the Board of Trade, various years (H.M.S.O., London).
Board of Trade, 1953-1956, Monopolies and Restrictive Practices Acts, 1948 and 1953:
Annual report by the Board of Trade, various years (H.M.S.O., London).
Bollen, K.A., Guilkey, D. K. and Mroz, T. A., 1995, Binary Outcomes and Endogenous
Explanatory Variables: Tests and Solutions with an Application to the Demand for
Contraceptive Use in Tunisia, Demography, 32, pp. 111-131.
Brock, W. A. and Scheinkman, J. A., 1985, Price setting supergames with capacity
constraints, Review of Economic Studies, 52, pp. 371-382.
Burn, D. (ed.), 1958, The Structure of British Industry, 2 volumes, (Cambridge University
Press, Cambridge).
Dick, A. R., 1996a, Identifying Contracts, Combinations and Conspiracies in Restraint of
Trade, Managerial and Decision Economics, 17, pp. 203-216.
Dick, A. R., 1996b, When Are Cartels Stable Contracts?, Journal of Law and Economics,
39, pp. 241-283.
Fershtman, C. and Pakes, A., 2000, A Dynamic Oligopoly with Collusion and Price Wars,
Rand Journal of Economics, pp. 207-236.
Fraas, A. G. and Greer, D. F., 1977, Market Structure and Price Collusion: An Empirical
Analysis, Journal of Industrial Economics, 26, pp. 21-44.
44
Hckner, J., 1994, Collusive Pricing in Markets for Vertically Differentiated Products,
International Journal of Industrial Organisation, 12, pp. 155-177.
Hart, P. E., Utton, M. A. and Walshe, G., 1973, Mergers and Concentration in British
Industry (Cambridge University Press, Cambridge).
Hay, G. A. and Kelley, D., 1974, An Empirical Survey of Price Fixing Conspiracies,
Journal of Law and Economics, 17, pp. 13-38.
Heath, J. B., 1961, Restrictive Practices and After, Manchester School of Economic and
Social Studies, 29, pp. 173-202.
Heath, J., 1963, Still Not Enough Competition?, 2nd edition (Institute of Economic Affairs,
London).
Hunter, A., 1966, Competition and the Law (George Allen, London).
Jacquemin, A., Nambu, T. and Dewez, I., 1981, A Dynamic Analysis of Export Cartels: The
Japanese Case, Economic Journal, 91, pp. 685-696.
Kitchin, P. D., 1976, Effective Rates of Protection for United Kingdom Manufacturing in
1963 and 1968, in Artis, M. J. and Nobay, A. R. (eds.), Essays in Economic Analysis
(Cambridge University Press, Cambridge).
Lydall, H. F., 1958, Aspects of Competition in Manufacturing Industry, Bulletin of the
Oxford University Institute of Statistics, 20, 319-337.
Lyons, B. and Matraves, C., 1996, Industrial Concentration, in Davies, S., Lyons B. et al.,
Industrial Concentration in the European Union (Clarendon Press, Oxford).
Maddala, G. S., 1983, Limited-dependent and Qualitative Variables in Econometrics
(Cambridge University Press, Cambridge).
Marquez, J., 1994, Life Expectancy of International Cartels: An Empirical Analysis,
Review of Industrial Organization, 9, pp. 331-341.
45
Martin, S., 2001, Advanced Industrial Economics, 2nd edition (Blackwell, Oxford).
Milward, A. S. and Brennan, G., 1996, Britains Place in the World: A Historical Inquiry into
Import Controls, 1945-1960 (Routledge, London).
Monopolies and Restrictive Practices Commission, 1955, Collective Discrimination: a
Report on Exclusive Dealing, Collective Boycotts, Aggregated Rebates and other
Discriminatory Trade Practices (H.M.S.O., London).
Monopolies and Restrictive Practices Commission, 1957, Report on the Supply and Exports
of Electrical and Allied Machinery and Plant (H.M.S.O., London).
Morgan, A. D. and Martin, D. 1975, Tariff Reductions and UK Imports of Manufactures:
1955-1971, National Institute Economic Review, 72, pp. 38-54.
O'Mahoney, M. and Oulton, N., 1990, Industry-Level Estimates of the Capital Stock in UK
Manufacturing, 1948-85, NIESR Discussion paper no. 172.
Phillips, A., 1972, An Econometric Study of Price-fixing, Market Structure and Performance
in British Industry in the early 1950s, in Cowling, K. (ed.), Market Structure and
Corporate Behaviour (Gray-Mills, London).
Political and Economic Planning, 1957, Industrial Trade Associations (George Allen &
Unwin, London).
Political and Economic Planning, 1959, Tariffs and Trade in Western Europe (George Allen
& Unwin, London).
Rivers, D. and Vuong, Q. H., 1988, Limited Information Estimators and Exogeneity Tests
for Simultaneous Probit Models, Journal of Econometrics, 39, pp. 347-366.
Scherer, F. M. and Ross, D., 1990, Industrial Market Structure and Economic Performance
(Houghton Mifflin, Boston).
Suslow, V. Y., 1991, Cartel Contract Duration: Empirical Evidence from International
46
47