Bidding competitiveness: empirical analysis from public construction contracts in Malaysia

BIDDING COMPETITIVENESS: EMPIRICAL ANALYSIS FROM PUBLIC CONSTRUCTION CONTRACTS IN MALAYSIA

RESEARCH MANAGEMENT UNIT UNIVERSITI TEKNOLOGI MARA SARAWAK 94300 KOTA SAMARAHAN SARAWAK, MALAYSIA

PREPARED BY

MOHD AZRAI BIN AZMAN

DECEMBER 2014

TABLE OF CONTENTS

LETTER OF REPORT SUBMISSION ........................................................................................iii LETTER OF OFFER ................................................................................................................. iv LETTER OF EXTENSION.......................................................................................................... v ACKNOWLEDGEMENTS ......................................................................................................... vi ENCHANCED RESEARCH TITLE AND OBJECTIVES ...........................................................vii PROPOSED EXECUTIVE SUMMARY - ORIGINAL PROPOSAL ............................................ 1 ENHANCED EXECUTIVE SUMMARY (ABSTRACT) ............................................................... 2 INTRODUCTION........................................................................................................................ 3 LITERATURE REVIEW.............................................................................................................. 5 METHODOLOGY ..................................................................................................................... 12 RESULTS AND DISCUSSION ................................................................................................ 19 CONCLUSION AND RECOMMENDATION ............................................................................ 28 REFERENCES ......................................................................................................................... 29 RESEARCH OUTCOMES ....................................................................................................... 32 APPENDIX 1: PRESENTER CERTIFICATE ........................................................................... 33 APPENDIX 2: ARTICLE ........................................................................................................... 34

ii

LETTER OF REPORT SUBMISSION

Tarikh

:

___Disember 2014

No. Fail Projek :

600-RMU/DANA 5/3 (4/2013)

PROF. MADYA DR. HJH. RASIDAH BINTI HJ. MAHDI Timbalan Rektor Penyelidikan dan Jaringan Industri UiTM SARAWAK

Prof. Madya,

PENYERAHAN

LAPORAN

COMPETITIVENESS:

DANA

EMPIRICAL

KECEMERLANGAN

ANALYSIS

FROM

2013

PUBLIC

-

BIDDING

CONSTRUCTION

CONTRACTS IN MALAYSIA

Perkara diatas adalah dirujuk.

Sukacita dimaklumkan, saya telah tamat membuat kajian. Dilampirkan bersama surat ini dua (2) laporan yang telah dijilid dan satu salinan ‘softcopy’ untuk pengesahan pihak Prof. Madya.

Perhatian Prof. Madya mengenai perkara di atas amat dihargai.

Sekian, terima kasih.

Yang benar,

_______________________ (MOHD AZRAI BIN AZMAN) No. UiTM: 299549

iii

LETTER OF OFFER

iv

LETTER OF EXTENSION

v

ACKNOWLEDGEMENTS

In the name of Allah, most gracious, most merciful

I would like to offer my sincere gratitude to Research Management Unit (RMU) and UiTM Sarawak for providing me the research fund.

I feel deeply indebted to Cawangan Kontrak dan Ukur Bahan (CKUB), Jabatan Kerja Raya Malaysia for giving me the permission to collect the unpublished bidding data.

On a more personal note, I want to acknowledge my deepest gratitude to my wife and daughter for their love and consideration. Lastly, to my parents and siblings who are away but always close to my heart.

vi

ENCHANCED RESEARCH TITLE AND OBJECTIVES

Original title as proposed: Bidding Competitiveness: Empirical Analysis from Public Construction Contracts In Malaysia

Improved/Enhanced title: As described above – no amendment.

Original objectives as proposed: 1. To identify significant factors which affect the bid spread distribution. 2. To categorize the bidding function of the data sets.

Improved/Enhanced title: 1. To relate the bidding competitive price in different bid auctions. 2. To analyze factors that affect bid ratios according to least square regression. 3. To contrast bidders’ bidding behaviour according to research results.

vii

PROPOSED EXECUTIVE SUMMARY - ORIGINAL PROPOSAL

The proposed research presents the empirical findings of bidding data collected from Public Works Department (PWD) of Malaysia. It attempts to recognize the bid competitiveness in respect of Average Bid Format used in the department. The study uses linear multiple regression analysis on project characteristics. The results could show that some project characteristics significantly affect the bid-spread (bid ratio). It may discover how bidding competitiveness can be explained through statistical analysis in terms of P-value and 2

adjusted R . The shape of probability curve will be studied using several statistical analyses to identify the bidding function. The findings could provide critical factors which affect the bid prices distribution. It could help PWD of Malaysia to reformulate the present use of Average Bid Format in PWD for current and future undertakings.

1

ENHANCED EXECUTIVE SUMMARY (ABSTRACT)

This research attempts to contrast bid competitiveness with respect to the average bid auction (ABA) and the non-ABA bidding formats used by the Public Works Department (PWD) of Malaysia. The research uses the ordinary least square regression and the Monte-Carlo simulation to point out significant predictors which affect the bid ratio and fitting probability distributions to bidding data, respectively. This research shows that the bidding strategy adopted is dependent on the different formats used. In the ABA format, bidders are more likely to submit identical bid prices. In the non-ABA format, they bid according to the first-price auction (FPA) strategy, which suggests greater variation between bid prices as a winning strategy and the reduction in the bid price to an estimated price ratio when more bidders bid. In addition, bidders lose more money when the distance between the project location and a firm’s operational office is greater. Best-fit probability density functions follow a gamma distribution for the ABA format and a Weibull distribution for the non-ABA format. The location and number of bidders affect bidders’ strategy to win. A competitive bidding price depends on a number of bidders who enter the auction. It increases the intensity of competition and it reduces the purchasing cost. The number of bidders is important because the cost of bidding may become a sunk cost. The probability to win an auction depends on the expected number of bidders. This affects bidders' decisions on whether to bid or not to bid. The research uses several statistical analyses to examine the 195 bidding data collected from Public Works Department of Malaysia. It finds that the project value, the distance from the project site to supply source and state region affect the number of bidders. This research presents empirical insights concerning the comparisons of different type of bidding formats, and its implications towards the construction client especially when it comes to its economic consequences.

Keywords: bidding, small and medium enterprises, public sector, tendering, average bid auction (ABA), first-price auction (FPA), competitiveness, construction; procurement; small; medium; contractor; microeconomics and Malaysia.

2

INTRODUCTION

A construction firm’s winning strategy for a bid is dependent on the auction format that requires a number of bidders to bid to encourage competition. Most auctioneers prefer inexpensive contracts through first-price auctions (FPAs). This auction format requires more than one bidder to bid at the lowest price (Drew & Skitmore, 1992). FPAs encourage bidders to reduce their bid price to the point at which they might lose money (Oo, Drew, & Runeson, 2010; Skitmore, Drew, & Ngai, 2001). Vickrey (1961) suggested that to overcome this problem, the contract should be awarded to the lowest bid but at the second lowest price. FPAs affect the auctioneer in the future when the actual contract price is revealed, which could lead to defaulting on the contract and expensive claims. FPAs affect the firm’s profits, which leads to the introduction of other bidding arrangements, such as the average bid auction (ABA) (Ioannou & Awwad, 2010; Ioannou & Leu, 1993). The ABA format uses the mean bid price to select the winning bidder.

Most bidders who bid for jobs related to the Public Works Department (PWD) prefer the ABA format because of the lower ex post risks involved (Halil, 2007). Utilizing the ABA format results in a firm gaining higher profits and a softening in the unnecessary low-priced bid (Ioannou & Awwad, 2010; Ioannou & Leu, 1993). The latter indicates that the bid to cost ratio is always higher when using the ABA format. However, a number of studies showed that this format encourages bidders to collude; thus, ABAs become less competitive compared with FPAs (Albano, Bianchi, & Spagnolo, 2006; Conley & Decarolis, 2012). Therefore, a need exists to enhance the auction format to ensure further improvement. An auctioneer needs to understand the consequences of the auction format implemented (Rothkopf & Harstad, 1994).

Research framework and problems The high-priced ABA format could reduce risks for the bidders and auctioneer. The ABA format could result in high profit margins for firms and lead to a reduction of claims and abandonment of projects. However, justification is required to ensure a fair price. It is a known fact that the lowest price will be awarded in FPAs. The ABA format becomes expensive if bidders collude, resulting in the auctioneer facing prices higher than the normal competitive price (Bajari & Summers, 2002).

Though the ABA format generates higher profits for potential bidders, winning more contracts is more important for firms’ survival in the business. The probability of winning a bid is lower when a very large number of firms operate in the market (Oo et al., 2010), because such a situation could lead to collusive behaviour among them. The conspiring bidders could strategically steer the threshold value of the mean to the amount desired by them (Conley & Decarolis, 2012).

3

This research analyzes the variables that affect bidding competition (i.e., whether additional bids reduce bid prices). The bid ratio (bid spread) could also be used to determine the bidding strategies of bidders. In addition, the ratio indicates the bidding function from which the data are drawn and may provide insights into how bidders’ strategize to win an auction. The following research questions are based on the research problems: 1. Could the number of bidders reduce the bid price? 2. How do the project characteristics affect the bid ratios according to a least square regression? 3. Could the bid ratios provide explanations for bidders’ bidding behaviour?

In addition, the objective of this research is to explore the factors that affect number of bidders in small and medium public procurement auction in Malaysia through empirical analysis on Public Works Department’s bidding data.

4

LITERATURE REVIEW

Background Public construction accounts almost 17% of total construction Gross Development Product (GDP) in Malaysia (CIDB, 2012). Since 2001, public construction contracted in huge percentage because of the economic liberalization agenda, as more investments for business facilities are made through private finances (Italian Trade Agency, 2012). Although the public finance construction has been reduced, the value of public construction is at RM13, 881 million. In Malaysia, there are too many contractors. The country has the highest builders’ licenses ownerships in the world (Adnan, Heap-Yih, Idris, & Ahmad, 2011). All procurement of works more than RM 500,000.00 must be contracted through private sources using open bidding to encourage competitive prices (Ministry of Finance, 2009). In terms of numbers, small and medium contractors account 93% or 56,904 from 61,292 of the total registered contractors (Ministry of Works, 2012). Table 1 shows Pusat Khidmat Kontraktor (PKK) registration of contractors according to bid values allowed and their paid-up capital. PKK is a public agency which facilitates the registration of indigenous contractors.

Table 1: Class of registration (Ministry of Works, 2012) Class of registration

Bid value [RM]

Paid-up capital [RM]

A

> 10 million

600,001

B

5,000,001 – 10 million

400,001

C

2,000,001 – 5 million

100,001

D

500,001 – 2 million

35,001

E

200,001 – 500,000

17,501

F

Below 200,000

10,000

PWD’s average bid auction The PWD of Malaysia uses ABAs and not FPAs to award less technical contracts to bidders. In FPAs, a bidder could win by discounting his bid price until he deems that he has the lowest bid (Ioannou & Awwad, 2010; Ioannou & Leu, 1993). Losses can be recovered through project management during the construction stage. ABA bid prices are high but could also reduce the risks of delay, substandard quality, and disputes (Ioannou & Awwad, 2010; Ioannou & Leu, 1993). In addition, the PWD uses the ABA format because the estimates prepared are not accurate enough for utilization as points of reference (PWD, 2004, 2005).

The PWD cost estimate and bid prices are used to obtain the cut-off price (COP). The bid price should not be lower than the COP. Bidders’ prices that are lower than 20% of the cost of building work, electrical work, and mechanical work and lower than 25% of the cost of civil engineering work from the mean of a builder’s work price are disqualified (PWD, 2005). The first step in the cut-off calculation is to exclude any bids that do not fall under the z-scale 5

group of –2.33 to 2.33 (normal distribution), which is the ideal range assumed by the PWD. Subsequently, the calculation of the cut-off price uses the lowest of the mean (1) and standard deviation (2): = −(

) ….….….….….….…. (1) and

= − ….….….….….….…. (2), where

represents the percentage according to project type.

Number of bidders and market conditions Theoretically, open bidding prepared on the basis of merit results in competitive bidding (Drew & Skitmore, 1992). Skitmore (1988) explained that selective bidding is also competitive when a reasonable number of bidders enter the auction. The number of bidders depends on economic conditions (Skitmore, 1988). In good economic conditions, when firms have abundant work, fewer bidders are interested in bidding; this ultimately increases bid prices. A study in Hong Kong showed that economic conditions as measured by the Tender Price Index (TPI) affect the number of bidders (Ngai, Drew, Lo, & Skitmore, 2002). Bidders in Hong Kong are more sensitive to economic conditions than bidders in Singapore even though they have almost identical characteristics. Bidders in Hong Kong are more likely to bid on projects when economic conditions worsen. However, bidders from Singapore are more likely to bid during good economic conditions; they are able to earn higher profits because most firms have full order books (Oo, Drew, & Lo, 2008).

Larger firms are more competitive in auctions for large contracts. Small- and medium-sized firms are obligated to opt for small- and medium-sized contracts, respectively (Drew & Skitmore, 1997). A study conducted in the United States found that when only one bidder enters an auction, the price is approximately 15% higher (deviation from the estimate) than when two bidders compete. This figure is reduced to 27% when eight bidders enter the auction and stagnates when more than eight bidders enter the auction (Carr, 2005). Because of the competitive nature of bidding, bidders must know their competitors. Bidders can obtain information on their competitors through personal contacts from suppliers and subcontractors (Oo et al., 2010). A strategic interest for most bidders is to conduct post-mortems after a client releases the list of bidders to identify active players in the market (Oo et al., 2008).

High-priced ABA bids reduce cost overruns in the future (Bucciol, Chillemi, & Palazzi, 2013), which occur only when a restricted number of bidders enter the bidding. An increase in the number of bidders increases the tendency of more bidders to collude. The experimental results in (Chang, Chen, & Salmon, 2013) showed that unlike in the FPA format, bidders in an ABA format tend to bid a higher price or a price equal to the true cost. Further, Chang et al. (2013) found no significant difference in the bidding behaviour for both auction formats. However, the experiment did not allow participants to coordinate. It is uncertain whether a price is competitive when more bidders enter a bidding, because price reduction could also 6

occur due to bidders’ desperation to win (Ngai et al., 2002). An increase in the number of bidders causes bidders to bid aggressively. When most firms become inefficient to operate, they embark on collusion, which in turn increases procurement costs (Estache & Iimi, 2011).

Large part of a bidding cost is un-coverable because it depends on the bidders’ initiative to bid and it could be recovered if they won the contract (Budde & Göx, 2000). In certain types of procurement i.e. traditional contract work in Commonwealth Countries, a public agency has to provide the costs of tender documents and consultations. Few costs are involved if the public agency chooses to use design and build construction. The public agency could only recover the documentation cost if more bidders submit bids.

Location The Malaysian economic environment is different compared with that of other nations. Per capita GDP is diverse among the states. The states in the southern part of the peninsula (Johor, Melaka, and Negeri Sembilan) are considered more developed and account for a larger portion of GDP per capita. Except for Penang, the states on the north and east coasts (Kedah, Kelantan, Terengganu, and Pahang) are less developed. Significantly, a larger portion of construction projects is concentrated in the southern part of the peninsular. A recent study showed that firms from Kelantan must bid on projects outside the state when only a few projects are available (Omran, Pakir, Rmali, & Termizi, 2011). Therefore, most firms prefer to bid near their area of operation. The federal government procedure entails that small projects (Class F) be awarded to local district firms (Ministry of Finance, 2009). Location plays an important role in determining the accuracy of cost estimations (Azman, Abdul-Samad, & Ismail, 2013). Location, which is public information, has been used as an important variable for measuring bid competitiveness and collusion (Bajari & Summers, 2002). Brockmann (2011) mentioned that it is unlikely for a construction contractor to place bid in an area which is far away from its operational office as this would increase its marginal cost e.g. transportation and logistics cost. In fact, a study shows that when it comes to supply-chains of construction products and methods, most contractors from rural areas need to find suitable suppliers in the urban area (Kamal & Flanagan, 2012). However, this brings to expensive cost of transportation for materials and machineries. Small contractors prefer to use local labour that lives near the area because they are cheaper than foreign labour. However, they are less productive and difficult to adapt new technology (Kamal & Flanagan, 2012).

Distribution of bid prices and ABA properties FPAs are always subjected to low bid prices (Herbsman & Ellis, 2006) and arise when a bidder who won a contract with a low bid price subcontracts the work, leading to quality degradation. The winning firm has to renegotiate with his subcontractors over their quoted price when the list of bidders is published (Dyer & Kagel, 1996). Italy, Taiwan, and some states in the United States used the ABA format to reduce unrealistically low bids (Ioannou & 7

Leu, 1993; Kumaraswamy & Walker, 1999). Recently, some Italian municipal and state departments decided to suspend the use of ABAs, because the format was proven to encourage the bidders to form cartels (Conley & Decarolis, 2012). Nevertheless, others contended the viability of this format, as it reduces “the winner’s curse.” ABAs are appropriate when bidders do not collude. Utilizing the ABA format improves the ex post revenue of the auctioneer (Chang et al., 2013).

The intensity of competition among bidders reduces the difference between the lowest and second lowest bids. In FPAs, the probability distribution function (PDF) is skewed (lognormal), which is near the lowest value because bidders attempt to concentrate on the lowest price (Skitmore et al., 2001). However, ABAs theoretically follow a normal distribution because bid prices concentrate near the mean value (Ioannou & Awwad, 2010; Ioannou & Leu, 1993). Bidding errors in FPAs are explained as bidders’ systematical reduction of the bid price to win the contract. Evidence shows that if bidders believe that a large number of bidders will enter a bid, they mark down the price, possibly resulting in a suboptimal price (Carr, 2005). In such circumstances, the winner of the contract experiences the so-called “winner’s curse” (Brockmann, 2011). Skitmore and Runeson (2006) found that the bidding behaviour of an individual bidder does not change over time.

The mistake in bidding is identifiable as the difference between bid amounts or the bid spread (Park & Chapin, 1992). In FPAs, the bid spread is the difference between the lowest and second lowest bids. The statistical properties of the bid spread have a number of potential uses, such as identifying mistakes in bids and determining the amount of bid security (Gates, 1961). Further, the bid spread is used to identify the distribution of bids. The best predictor variable of the spread is the percentage difference between the expected value of the lowest order statistic and the expected value of the second lowest order statistic (%λ) of the lognormal distribution (Skitmore et al., 2001; Skitmore & Lo, 2002). Further, the bid spread can be used to detect the phenomena in non-traditional auctions (Conley & Decarolis, 2012; S. Li, Foulger, & Philips, 2008). The spread is always lower in ABAs when more bidders enter the bidding (Conley & Decarolis, 2012). The same result is found for FPAs when more bidders enter the bidding, thus reducing the spread (S. Li et al., 2008). In addition, the contract value also reduces the spread, which could also happen because bidders may be attracted to the large value of potential projects (Skitmore et al., 2001).

The homogeneity assumption predicts the probability of winning a bid when the density function of winning a bid is known. According to Skitmore (2001), for the sake of simplicity, each bidder is assumed to bid from the same distribution. Beeston (1983) acknowledged that the degree of the skew could indicate the level of bidders’ competitiveness. A positive skew indicates greater competitiveness. If stationary is assumed, the PDF of bids has fixed parameter values (Skitmore & Runeson, 2006). However, the nature of construction, such as 8

changes in workloads and market conditions, affects bidders’ strategies over time (Oo et al., 2008). Skitmore (1988) indicated that out of 29 studies, 9 are normal, 7 are lognormal, 4 are uniform, 3 are gamma, 3 are positively skewed, and 3 come from other distributions. Beeston (1983) suggested that competitive bids are almost symmetrical for practical purposes, because the skewness is small and can be ignored. Fitting a PDF to bidding data is complicated because of the special characteristics of the bidding data, such as the differences in project values, number of bidders, and auctions (Skitmore, 2013). In addition, the bidding data are affected by local procedures for bidding arrangements and the intensity of the market conditions.

Bidding environment and adverse selection process Malaysia has many construction firms. The country is said to have one of the highest ownerships of builders’ licenses (Adnan et al., 2011). The Malaysian government promotes the participation of indigenous (bumiputera) people in the economy through licenses (Rasiah & Shari, 2001). The participation of government-linked companies (GLCs) in public construction contracts also may have contributed to the impending competition, because these companies are established in a way similar to private firms (Abdul-Aziz, Jaafar, & Hussin, 2007). A new study by (Menon & Ng, 2013) showed that additional investments by GLCs create a crowding out effect on private investments. Most government contracts, except for critical jobs, are allocated to the indigenous population. In addition, such contracts are widely considered difficult-to-win main contract and subcontract works in private sectors and are overly dependent on the patronage system (McCrudden & Gross, 2006). Before 2001, the percentage of public projects in terms of value exceeded those of private sector and GLC projects. However, Figure 1 indicates that the percentages of private sector and GLC projects in construction are now larger than that of public sector projects (CIDB, 2012). Unhealthy competition has resulted in a government policy that does not require a prequalification procedure for small- and medium-sized contracts. Prequalification is only required for contracts of large value.

9

90% 80% 70% Value of public project in %

60% 50%

Value of private project in %

40% 30%

No. of public project in %

20% No. of private project in %

10% 2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

0%

Figure 1: Percentages of public versus private projects in terms of value and number of projects in Malaysia (2000-2012) (abstracted from: CIDB, 2012)

Changes in the economic spectrum could encourage rent-seeking behaviour. In addition, the problem with ABAs is that the probability of winning a bid is much lower than in FPAs (Ioannou & Awwad, 2010; Ioannou & Leu, 1993). This problem may convince bidders to engage in potential collusive behaviour to rig the bid. Therefore, subjecting the bidding process to a number of “dummy” bidders is predicted to improve the chances of winning contracts. Based on theoretical predictions, bidders have incentives to submit identical bids, leading to a Nash equilibrium wherein most bidders submit the same bid prices to ensure their profitability. To overcome this problem, the collusive bidders price their bids more closely to one another (Conley & Decarolis, 2012).

Recently, a directive letter from the PWD’s director accepted that ABAs could be manipulated by bidders in the form of a cartel (PWD, 2012). Therefore, the PWD publicized the department’s estimates such as reserve prices. Therefore, the problem of some firms gaining an upper hand in the format is believed to be eliminated. Conley and Decarolis (2012) explained that most bids are in the range of the winning discount, indicating that the probability of winning is low. To break the tie, colluding bidders submit bids that are placed not to win but to pilot the average. The bids are often priced with high and low discounts. Subsequently, the trimmed mean is calculated to ensure a single cartel member bid price in a desired winning range. Chang et al. (2013) disagreed with the view that ABAs are high priced, because they found that ABA prices are slightly higher compared with those in FPAs but not significant. However, they agreed that ABAs could be comparatively more collusive.

10

Bidders’ demographic and public information Bidders evaluate two types of information before placing bids: private information and public information. Private information is the bidder’s estimate of the project value; bidders will not have common estimate values. Public information is the information known to all bidders, such as the number of potential competitors, locality, economic conditions, and availability of potential projects. This information can be freely obtained. The construction firms considered in this paper are small and medium sized. Some firms from other states were exempted, as they had insufficient data. All firms specialize in completing reinforced concrete building work. Table 2 shows the number of potential competitors based on class. Terengganu has the highest and Penang has the lowest number of firms. In addition, Table 1 shows demographic information on the selected states in Peninsular Malaysia. Johor has the highest number and value of projects awarded, whereas Kelantan has the lowest number and value of projects awarded.

Table 2: Number of indigenous contractors and demographic information for selected states (abstracted from: CIDB, 2012; Department of Statistics, 2013; PKK, 2013) Variable Class A

Johor

Kedah

2

K’tan

4

80”

72”

3

Melaka

3

76”

5

N9

9

29”

2

Pahang

7

6

38”

9

5

59”

8

61”

6

41”

Class C Class D

189”

96”

159”

130”

146”

188”

Class E

182”1

81”8

91”7

92”6

126”4

Total

600”2

348”7

399”6

52”

108”2

64”7

101”3

2

8

5

1

11”

36”9

87”4 6

298”8

4

32”

67”5 7

479”4

3

22”

9

3

8

67”5

9

87”

183”

231”1

121”5

54”9

140”2

127”3

487”3

276”9

462”5

695”1

5

2

19,210”3

C

5

17,802”

11,657”

6,821”

22,288”

24,233”

19,111”

32,376”

D

8,720”1

4,091”3

1,503”9

2,076”8

5,712”2

3,143”7

E

1,070”1

409”2

162”9

243”8

306”7

432”5

9,500”6

15,099”4 8

9

1,664”8

1.48”

6,686”7 3

1.54”

36,137”1 2

152”1 4

6

0.98”

78”1

45”8

B

0.74”

107”1

4

36”

3.19”

1.58”

T’ganu

8

29”

A

1.89”

Perak 36”

7

Class B

35”

Penang

4

1.01”7

2.36”

1,048”9

21,035”2

13,035”5

7

14,703”

15,883”6

4,069”4

3,681”6

3,783”5

435”4

501”3

377”6

1

” refers to ranking. K’tan = Kelantan, N9 = Negeri Sembilan, and T’ganu = Terengganu. “A” for population (million), “B” for area (km2), “C” for GDP (RM current price), “D” for total project value (RM million), and “E” for number of projects awarded. “Class” indicates type of contracting firms.

11

METHODOLOGY

The analysis uses 195 data sets representing school building projects collected from the PWD of Malaysia. This type of data is acceptable for research because it is homogeneous and repetitious, making such data easy to predict and act as the main element for a simple form of collusion (Brockmann, 2011). Drew and Skitmore (1997) emphasized that competiveness in bidding can be modeled by analyzing the entire distribution of bids, competitiveness within bids, and competitiveness between bids. The research highlights the competiveness between bids given the nature of the data collected and is concerned with the bidding performance of bidders competing with one another and the entire bid distribution. The data are limited to class contractors from B to E class which are considered small and medium class contractors according to PKK registration. All projects were restricted to the first half of 2007 and located in Peninsular Malaysia.

Selection of variables of the model A bid is priced on the basis of project characteristics, including the physical appearance and contractual arrangement of contract bidding. In previous studies, the size and number of bidders were considered the important variables to explain systematic deviations between bids (Skitmore et al., 2001). However, many unknown factors could be recovered using the number of variables studied, such as project value in MYR (Malaysian Ringgit), states in the Peninsular of Malaysia (eight), distance of the project from the Central Business District (in kilometers), number of bidders, class of the construction firm (Class B, C, D, and E), and economic factors of the district (rural and urban).

Bid ratio In FPAs, the ratio between the first and second bid is used to show the ratio of the first bid’s underbidding of the second bid to win the auction. However, to use the average bid in PWD, some adjustments are required to analyze the different type of bidding arrangement used. A bid closer to the cutoff price is the lowest bid to be awarded. In FPAs, the ratio is calculated according to the amount by which the first bid underbids the second lowest bid. The calculations of the bid ratio distributions are as follows:





(



)=



−







The difference between the lowest and second lowest bid indicates the mistakes in bids for FPAs.



(



)=



12

−





The winning bid ratio (WBR) shows the spread of the mean of the bids and the second nearest bids, indicating the mistake in bids for ABAs. The following ratios are divided by the number of bidders ( − 1) for the average ratios:









)=(

( (





−





)=(

)/( − 1)

−







)/( − 1)

MeBR and MMBR show the degree of clustering among bids. A large ratio indicates that greater difference between bids and vice versa (Williams, 2005). Further, a large ratio detects clustering of bids near the winning region for ABA, which is near the mean of the bids.



(



)=(

−



)/( − 1)

The ratio shows the spread of the bids and whether a significant variation exists in the range of bid values. A larger ratio indicates that a number of bids are placed higher than the cheapest bid as well as the intensity of the clustering around the bid prices.

Competitiveness of bid prices The regression of the mean of the bids’ deviation from the estimate versus the number of bidders suggests bid competitiveness in terms of price reductions. Figure 2 shows that an additional number of bidders significantly reduces the bid price regardless of the different bid formats used (F = 3.902, df1 = 1, df2 = 165, p < 0.05) according to a logarithmic function. However, the ANOVA results show no significant relationship (p > 0.05). To differentiate the bid formats used, the analysis focuses on two groups—less than 10 bidders (non-ABA) and 10 bidders or more (ABA).

The analysis shows that when the non-ABA format is used, the bid price is significantly reduced according to a linear (F = 8.123, df1 = 1, df2 = 36, p < 0.05) and a logarithmic (F = 11.569, df1 = 1, df2 = 36, p < 0.05) function, respectively. ANOVA results also indicate that 2

the relationship is significant (F = 8.123, df = 37, p < 0.05) with an adjusted R of 16%. However, bid price reduction is not significant (p > 0.05) when the ABA format is used; ANOVA shows no significant relationship (p > 0.05).

13

Figure 2: Natural log mean of bids higher than the estimate versus number of bidders for all bid formats (above) and non-ABA format (below)

Linear multiple regression on bid ratios and project characteristics Table 3 provides an initial analysis of the bid ratios calculated using different bid formats. The data collected transform into five different bid ratios: SLBR, WBR, MeBR, MMBR, and MaBR. The market price does not require adjustment because all bid prices originated from the first half of 2007. To differentiate bidding strategies in different bidding auctions, the analysis is separated into two groups: 10 bidders or more (ABA format) and less than 10 bidders (nonABA). 14

Table 3: Initial data analysis on datasets Item

ABA

Non-ABA

151

44

No. of auctions Total bidders

2711

310

Min. no. bidders

10

3

Max. no. of bidders

50

9

17.95

7.04

Mean no. of bidders Std. Dev. of no. of bidders

6.42

1.75

Min. contract value (RM)

296,866

217,262

Max. contract value (RM)

7,720,120

11,115,264

3,095,320.32

2,900,404.61

Mean contract value (RM)

Linear multiple regression (LMR) predicts the linear interactions between dependent and independent variables given that the dependent variable may be affected by more variables (H. Li, Shen, & Love, 2005; Ping Yung, 2010). The stepwise technique is used to select predictor variables. LMR requires an a priori assumption regarding the relationship among variables in a set of equations as follows: =

where

+

ln ( ) +

( ) … … … … …

is the natural logarithm; ln

natural logarithm.

,

, and

(

)+ε,

is the natural logarithm of the outcome bid ratio; for

are predictors of project characteristics;

the parameters calculated using the least square approach; and

,

,

and

are

is the random error of .

Field (2009) suggested that a sample size of 200 with 20 predictors is enough to find a moderate effect size. The following calculation uses G*Power software to determine the sample size for LMR (Faul, Erdfelder, Lang, & Buchner, 2007): = 23.55 = (0.15) x N Estimated total sample size (N) = 157 Non-centrality parameter λ

= 23.550

Critical F

= 1.648

Numerator df

= 20

Denominator df

= 136

Actual power (1 – β)

= 0.803

where 1 – β is the power of the test, and β is the probability of falsely accepting H0 when in fact H1 is true. Power analysis uses a moderate effect size ( = 0.15) to find an acceptable interpretation and reject the Type II error.

is Cohen’s effect size. Based on this power

analysis, a minimum sample size of 157 is needed. Tables 4 and 5 show the LMR test results on ABA and non-ABA bid ratios.

15

Table 4: LMR test results on bid ratios (ABA) Predictor

SLBR

MeBR

MMBR

MaBR

WBR

Penang

-

0.041 (2.064)*

-

0.022 (2.319)*

-

N9

-

0.006 (2.794)*

0.013 (2.510)*

0.001 (3.286)*

-

0.005 (−2.834)*

-

-

-

-

-

0.036 (−2.115)*

0.001 (−3.290)*

0.006 (−2.820)*

-

T’ganu & K’tan

0.011 (−2.560)*

-

-

-

0.018 (−2.395)*

No. of bidders

-

0.000 (−10.664)*

0.000 (−4.427)*

0.000 (−7.218)*

-

Class C

-

-

0.023 (−2.305)*

-

-

-

-

0.000 (−3.829)*

0.006 (−2.791)*

-

Johor Pahang

Class E 2

Adjusted R

0.069

0.449

0.226

0.337

0.032

F-statistic

6.517

31.505

8.896

14.734

5.736

Syx

1.658

0.445

0.513

0.425

1.224

Model P-value

0.002

0.000

0.000

0.000

0.018

VIF

1.015

1.027

1.047

1.067

1.000

Random

Random

Random

Random

Random

Condition index

1.721

18.681

21.956

19.361

1.252

Shapiro–Wilk on

Statistic = 0.902, df = 151,

Statistic = 0.996, df = 151,

Statistic = 0.988, df = 136,

Statistic = 0.991, df = 136,

Statistic = 0.923, df = 143,

Residuals

residuals

P = 0.000

P = 0.946

P = 0.298

P = 0.517

P = 0.000

Case-wise

6 cases (3.97%)

10 cases (6.62%)

4 cases (2.94%)

4 cases (2.94%)

5 cases (3.50%)

diagnostics (-) Statistically insignificant when P > 0.05. (*) Statistically significant when p < 0.05. Values in brackets are the t-values. Syx is the standard error of estimates.

16

Table 5: LMR test results on bid ratios (non-ABA) Predictor

SLBR

MeBR

MMBR

MaBR

WBR

0.031 (2.233)*

-

-

-

-

T’ganu & K’tan

-

-

0.005 (−2.955)*

0.031 (−2.240)*

0.002 (−3.351)*

No. of bidders

-

0.041 (−2.103)*

0.004 (−3.067)*

0.008 (−2.798)*

-

Distance

-

-

-

-

0.005 (2.993)*

Adjusted R

0.085

0.074

0.304

0.228

0.249

F-statistic

4.987

4.423

10.381

7.332

7.959

Syx

1.336

0.464

0.472

0.426

1.226

Model P-value

0.031

0.041

0.000

0.002

0.001

VIF

1.000

1.00

1.016

1.016

1.079

Random

Random

Random

Random

Random

Condition index

1.732

13.694

14.089

14.089

12.404

Shapiro–Wilk on

Statistic = 0.902, df = 44,

Statistic = 0.967, df = 44,

Statistic = 0.969, df = 44,

Statistic = 0.971, df = 44,

Statistic = 0.933, df = 43,

residuals

P = 0.000

P = 0.228

P = 0.268

P = 0.335

P = 0.014

Case-wise

2 cases (4.54%)

2 cases (4.54%)

2 cases (4.54%)

3 cases (6.81%)

2 cases (4.65%)

N9

2

Residuals

diagnostics (-) Statistically insignificant when P > 0.05. (*) Statistically significant when p < 0.05. Values in brackets are the t-values. Syx is the standard error of estimates.

17

Fitting bid distributions The Kolmogorov-Smirnov (K-S) test compares the distribution of the empirical bids and assesses whether they belong to the same distribution. The Monte Carlo (M-C) simulation predicts the sampling distribution of empirical statistics and simulates the bid distribution of the empirical results using a stochastic process (Mooney, 1997). The M-C simulation generates a random set of possible bid ratio values according to a parameter estimation based on the maximum likelihood format. Two solutions exist for the parameter estimation: the closed-form and the Newton-Rhapson (N-R) formats. Closed-form formats are used in many types of distributions. The N-R format is used in distributions such as binominal, beta, gamma, and Weibull distributions. This approach requires a log-likelihood function, the gradient vector, the Hessian matrix, and initial values from the N-R process. The simulation creates distributions that fit the empirical bid ratios through a set of 100,000 cases using a random number generator. The modified Anderson-Darling (M-A-D) goodness of fit determines the shape of the best-fit PDF. The unmodified Anderson-Darling (U-A-D) is defined as follows:

H0: The data follow a specific distribution. Ha: The data do not follow a specific distribution. The U-A-D test statistic: =− − =∑ where

(

)

[ ( )+

1− (

) ],

is the cumulative distribution function, and

is the order of statistics.

The critical values (significance α level and critical region) for the U-A-D are in accordance with values and formulas for specific distributions. The test is one-sided, and the hypothesis is rejected if the test statistic

is greater than the critical value (α = 0.05). M-A-D with a weight

function is more robust than U-A-D but comprises more complicated calculations (Shin, Jung, Jeong, & Heo, 2012). However, M-A-D applies the same principle. The test statistic uses MA-D with a frequency weight. For more information on M-C and the M-A-D goodness-of-fit test, see D’Agostino and Stephens (1986), Johnson et al. (2005), Kotz and Van Dorp (2004), Kroese et al. (2011), and Marsaglia and Marsaglia (2004).

If the result from the empirical analysis fits the simulated data, the raw data might be derived from the same distribution. The analysis uses the bid ratio based on an adjusted multiplicative absolute number because the winning bid is used as a starting point; the second nearest bid could be less or more than the winning bid based on the ABA mechanism. The formula is as follows: |

|





(

)| = ( 18

)| =





−1

−1

RESULTS AND DISCUSSION

First, to differentiate the distributions of bids, it is better to compare AWBR and AFPR in ABA and non-ABA bids, respectively. Figure 3 shows the empirical distributions and Figure 4 shows the simulated distributions. Table 6 shows the descriptive statistics of the empirical and simulated distributions based on ABA and non-ABA auctions. Shapiro-Wilk rejects it, because it is normally distributed in both ratios (p < 0.05). The empirical distribution is shown to be asymmetric with a longer right tail, indicating that the distribution is skewed to the right. Table 7 shows the ranking according to a descending order of empirical distributions based on Anderson-Darling (A-D) fit statistics. The two-tailed K-S test, which compares whether the bid ratio comes from the same distribution, indicates that the AWBR of ABAs and non-ABAs does not come from the same distribution (P < 0.01, Statistic = 3.286) at a 99% confidence interval. Further, in AFPR, no differences exist between distributions for ABAs and non-ABAs at the 99% confident interval (p > 0.01, Statistic = 0.701).

Table 6: Descriptive statistics on empirical and simulated distributions of bids in different auctions Empirical bid ratio ABA - AWBR

Simulated bid ratio

Non-ABA

-

ABA - AWBR

AFPR Statistic

ABA

non-

Non-ABA

-

AFPR

ABA

non-

ABA

ABA

ABA

non-

ABA

ABA

nonABA

Mean

0.014

0.027

0.065

0.061

0.014

0.027

0.065

0.061

Median

0.009

0.017

0.038

0.037

0.010

0.017

0.040

0.043

Std. Dev

0.013

0.029

0.072

0.070

0.013

0.030

0.073

0.061

Coeff.

0.929

1.074

1.108

1.148

0.962

1.111

1.123

1.000

Skewness

1.648

1.469

1.976

2.873

1.915

2.144

2.247

1.983

Kurtosis

2.651

1.560

4.389

11.118

5.489

9.090

7.560

5.883

Variation

19

Figure 3: Histogram of empirical winning bid (top) and FPA (bottom) ratios Table 7: Best-fit statistics of empirical distributions based on bid ratios in descending order Ratio

ABA - AWBR

Non-ABA - AWBR

ABA - AFPR

Non-ABA - AFPR

Best fit

Fit statistics (A-D)

Parameters

Gamma

statistic = 0.51, p = 0.21

θ = 77.61, k = 1.07

Weibull

statistic = 0.54, p = 0.18

λ = 0.01, k = 1.04

Exponential

statistic = 0.61, p = 0.36

λ = 72.43

Weibull

statistic = 0.22, p = 0.25

λ = 0.03, k = 0.88

Gamma

statistic = 0.24, p = 0.25

θ = 30.76, k = 0.82

Lognormal

statistic = 0.48, p = 0.22

μ = 0.01, σ = 1.38

Exponential

statistic = 0.74, p = 0.24

λ = 37.43

Gamma

statistic = 0.17, p = 0.25

θ = 12.05, k = 0.78

Weibull

statistic = 0.18, p = 0.25

λ = 0.06, k = 0.86

Exponential

statistic = 0.40, p = 0.64

λ = 16.33

Gamma

statistic = 0.41, p = 0.25

θ = 15.87, k = 0.97

Weibull

statistic = 0.41, p = 0.25

λ = 0.06, k = 0.97

20

The M-C method simulates the ratios according to uniform, normal, and lognormal distributions to recognize the common distributions from which the data could be derived. The A-D test shows that non-ABA-AWBR simulated data are fitted only to the lognormal (p > 0.05) and rejects that ABA-AWBR simulated data are derived from those distributions. The M-C method simulates the empirical data according to gamma, Weibull, exponential, and lognormal distributions (refer to Table 7). ABA-AWBR shows that gamma- and Weibullsimulated data fits gamma and Weibull distributions, respectively but also that exponential simulated data fits Weibull, gamma, and exponential distributions. In non-ABA-AWBR, the simulated data are shown to fit with their respective distributions except for exponential data, which fits the Weibull distribution. In ABA-AFPR, gamma and Weibull data fit only gamma and Weibull distributions, respectively. In non-ABA-AFPR, exponential data fits exponential, Weibull, and gamma. Gamma data fits gamma and Weibull distributions. Weibull data fits Weibull and gamma distributions. The simulated PDF of non-ABA-AFPR shows that exponential data is the best fit. However, this fit may also result from Weibull and gamma data, which could become exponential when k ≥ 1.

Table 7: Best-fit statistics of simulated distributions based on bid ratios in descending order Ratio

Simulated

Best fit

Fit statistics (A-D)

Parameters

Gamma

Gamma

statistic = 0.29, p =

θ = 77.85, k = 1.07

0.25 ABA - AWBR

Weibull

Weibull

statistic = 0.25, p =

λ = 0.01, k = 1.04

0.25 Exponential

Weibull

statistic = 0.25, p =

λ = 0.01, k = 1.00

Gamma

0.25

θ = 72.66, k = 1.00

Exponential

statistic = 0.29, p =

λ = 72.50

0.25 statistic = 0.44, p = 0.57 Weibull Non-ABA AWBR

Weibull

statistic = 0.25, p =

-

λ = 0.03, k = 0.88

0.25 Gamma

Gamma

statistic = 0.30, p =

θ = 30.86, k = 0.82

0.25 Lognormal

Lognormal

statistic = 0.28, p =

μ = 0.01, σ = 1.38

0.65 Exponential

Weibull

statistic = 0.25, p =

λ = 0.03, k = 0.88

0.25 ABA - AFPR

Gamma

Gamma

statistic = 0.30, p =

θ = 12.10, k = 0.78

0.25 Weibull

Weibull

statistic = 0.25, p = 0.25 21

λ = 0.06, k = 0.86

Exponential

Non-ABA

-

Weibull

statistic = 0.25, p =

λ = 0.06, k = 1.00

Gamma

0.25

θ = 16.38, k = 1.00

Exponential

statistic = 0.29, p =

λ = 16.35

AFPR

0.25 statistic = 0.44, p = 0.57 Gamma

Gamma

statistic = 0.30, p =

θ = 15.92, k = 0.97

Weibull

0.25

λ = 0.06, k = 0.98

statistic = 0.62, p = 0.12 Weibull

Weibull

statistic = 0.25, p =

λ = 0.06, k = 0.97

Gamma

0.25

θ = 15.59, k = 0.95

statistic = 0.69, p = 0.09

Figure 4: Histogram of simulated winning bid ratio (top) and FPA ratio (bottom) of ABA and non-ABA 22

Only three (3) variables are selected for the analysis i.e. state (of Kedah, Penang, Perak, Negeri Sembilan, Melaka, Johor, Pahang, Terengganu and Kelantan), project estimation in Ringgit Malaysia [RM] and the distance from project site to Central Business District (CBD) in Kilometre [km]. Natural logarithm transformation improves the linearity of the data (denoted as ln or LN). Analysis shows that 165 or 89.67% out of 184 bidders had won the contracts to carry out construction businesses in their respective states. The following Table 8 is the descriptive statistics of data collected for selected states in Peninsular Malaysia. It shows that Terengganu and Kelantan have the highest number of bidders in terms of mean with 25 bidders per auction and Johor has the lowest with 11 numbers of bidders per auction. Pearson’s coefficient shows that the correlation between project estimation and distance from site to CBD is not significant according to 2-tailed test (p>0.05). However, it finds that the relationship between distance from site to CBD and state is significant (p