230 likes | 371 Views
Sealed-Bid Auctions: Theory and Application. Frans van Schaik & Jack Kleijnen September 2001 Slides and paper: see web page ( Web: http://www.tilburguniversity.nl/faculties/few/im/staff/kleijnen/ ). Overview. Problem: Sealed bid auction: case study of mussel auction (Yerseke)
E N D
Sealed-Bid Auctions: Theory and Application Frans van Schaik & Jack Kleijnen September 2001 Slides and paper: see web page (Web: http://www.tilburguniversity.nl/faculties/few/im/staff/kleijnen/)
Overview • Problem:Sealed bid auction: case study of mussel auction (Yerseke) • Method:Regression analysis of 28,017 mussel lots • Results:Reject four null-hypotheses: for example,big buyers do not pay more per mussel ton!Novel tool for buyer’s performance evaluation‘Hedonic’ price factors: measure objective characteristics Van Schaik & Kleijnen: Sealed-bid auctions
Auctions and supply chain management Hypotheses:1. Should big buyers avoid auctions? 2. Does large supply decrease prices? 3. Is seasonality important? 4. Does the buyer perform well? Model: ‘hedonic’ price Van Schaik & Kleijnen: Sealed-bid auctions
Dutch mussel industry • Origin: Eastern Scheldt (Oosterschelde) and Wadden Sea • 80 ships (farmers) • Traded per lot (usually a ship) • Large differences in quality of mussels • Highly perishable Van Schaik & Kleijnen: Sealed-bid auctions
Usage: live versus processed mussels live processed mussels musselsshare at auction 84% 16% quality needed high low number of buyers 20 4 buying costs as % of turnover 62% 45% Van Schaik & Kleijnen: Sealed-bid auctions
Importance of purchasing Van Schaik & Kleijnen: Sealed-bid auctions
Yerseke mussel auction • Quality control by auctioneer, per lot • Net weight = gross weight - impurity (starfish etc.) • Data over 14 years: 1986/1987 through 1999/2000 Van Schaik & Kleijnen: Sealed-bid auctions
Mussel auction design • Sealed-bid or simultaneous single-shotPro: Fast (30 lots per hour)Con: Less information on other buyers • Random sequence of lots, offered at auction Van Schaik & Kleijnen: Sealed-bid auctions
Auction: buyer’s problems • Expensive trade: 1 lot may cost € 140,000 • Lots are heterogeneous: many types and sizesConsumer is offered 5 sizes only • Lots are sold sequentially:desired market share realized? Van Schaik & Kleijnen: Sealed-bid auctions
Regression model Linear model: y = b0 + b1x1 + b2x2 +… + e withy = lnpso(i) y = (ln p) = p / p: relative price (no scale effect: € or f)(ii) If x = ln z (double-log transform),then b is elasticity coefficient ( p / p)/ ( z / z) But: single-log transform x8 = z8 so ( p / p) = b8 z8(also see scatter plot) Moreover: binary variables; example:z11 =1: Wadden origin; z11 =0: Scheldt origin [ln(0) undefined!]Wadden mussels are b11 % more expensive: p/ p = exp(b11) - 1 e: either white noise (NIID) or auto-correlated noise Van Schaik & Kleijnen: Sealed-bid auctions
Regression model: independent variables of type 1 4 hypotheses: 1. market share z1 2. annual supply z2 3. seasonality z3 4. buyer’s performance z4 Van Schaik & Kleijnen: Sealed-bid auctions
Regression model: independent variables of type 2 8 concomitant (control) variables: • Hedonic price factors (auction supplies info sheet):Meat yield z5 (meaty mussels are expensive)Size count z6 (smaller mussels are cheaper)Impurity z7Barnacles z8 (‘pokken’)Slippers z9 • Usagez10: live mussels (expensive) versus processed mussels • Originz11: Wadden (expensive) versus Scheldt • Trendz12: prices increased over the years Problem: collinearities between origin z11, meat z5, size z6resp. season z3, meat z5, size z6 Van Schaik & Kleijnen: Sealed-bid auctions
Data source and selection • Computerized auction, since 198630,303 lots; after deleting ‘irregular’ data 28,017 lotsExample: 1999: 2,199 lots • Measured per lot: hedonic variables z5 through z11 (plus z4, buyer) • Measured per year (14 years): share z1, supply z2, trend z12 • Measured in days (1, 2, …): season z3 Van Schaik & Kleijnen: Sealed-bid auctions
Regression results White noise assumption: OLS gives BLUEAll t-statistics significant: many data availableBut: Durbin-Watson is significant (0.954): auto-correlated noise! GLS with 1st order auto-correlation : Prais-Winsten method (SPSS)Estimate of : 0.6Durbin-Watson is not significant (2.290) Relative effects: standardize effects (b var(z)1/2) Most important effects: supply, size, season, trend, meat Adjusted R2: 56% (66% for OLS); between 37% and 93% in refs. Van Schaik & Kleijnen: Sealed-bid auctions
Regression results: 4 hypotheses All 4 hypotheses rejected; signs correct: • Big buyers pay more • Supply z2 has elasticity coefficient -0.8 • Season starts expensive • Some buyers perform better Van Schaik & Kleijnen: Sealed-bid auctions
Regression results: control variables Quality variables z5 through z9 have expected signs Origin: Wadden has negative effect!Practice: Wadden mussels are expensiveExplanation: Wadden mussels are meaty, big, and early in seasonBut: Wadden mussels travel much (5% death) Note on robustness: Results per year agree with results for total sample (14 years) Van Schaik & Kleijnen: Sealed-bid auctions
Buyer’s performance: pays 6% premium explainsthis buyer has large market share 1%he is a live-mussel trader 1%he buys good quality (hedonic factors) 14%large annual supply -2%trend: he started late 11%season: live-mussel traders must buy early 3% total effect 31%but: actually he paid 37%premium paid 6% Van Schaik & Kleijnen: Sealed-bid auctions
Out-of-sample prediction • Drop most recent mussel year (1999) • Re-estimate from 1986-1998 • Predict 1999 (2,199 lots) Predicted prices show less variation than actual prices (R2 < 1) Van Schaik & Kleijnen: Sealed-bid auctions
Prediction with autocorrelation Weights of older prediction errors decrease geometrically Sum of geometrical weights equals 1 Van Schaik & Kleijnen: Sealed-bid auctions
Predictive power OLS GLS Van Schaik & Kleijnen: Sealed-bid auctions
Buyer’s decision support system Goal: Buy (say) 20% of all lots in 1999 1986-1998: 20% quantile of actual price - predicted price: -241999: bid = predicted price minus 24 Result:18% market share Resembles Black-Scholes option model(regression model versus differential equations ) Van Schaik & Kleijnen: Sealed-bid auctions
Conclusions • Regression model uses many data (28,017 lots) • Four null-hypotheses rejected • Surprise: negative Wadden effect • Novel tool for buyer’s performance evaluation Van Schaik & Kleijnen: Sealed-bid auctions
Limitations and future research Limitations: • Only successful bids recorded • Dutch mussel auction versus building, UMTS, etc. • Not recorded: color, taste, texture Future research: • Spot versus futures market (shrimps in Minneapolis) • Improving auction design Van Schaik & Kleijnen: Sealed-bid auctions