A General Equilibrium Model of Supply Chain Interactions and Risk
Transcription
A General Equilibrium Model of Supply Chain Interactions and Risk
A General Equilibrium Model of Supply Chain Interactions and Risk Propagation John Birge and Jing Wu1 1 University of Chicago Booth School of Business SCF Symposium, Madrid John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 1 / 29 S&P 500 Supply Chain Network ACT ACT HP HP SXL SXL FLR ADI FLR NEM ADI MMM NEM MMM AN CNP ALV CREE KO GAS EIX AN BDX LEA CERN AEP BDX LEA CERN EIX CTSH NXPI EMR CMI BWP CNP ALV CREE KO GAS CTSH NXPI EMR CMI AEP BWP ESRX ESRX PRGO PRGO KR CSX MUR KR CSX MUR BWA ADBE CRM DISCA ITW BWA ADBE CRM DISCA ITW ABT ABT AMGN AMGN RAI RAI CBS CBS CTXS GLW CTXS GLW SNI IEP SNI RAX TSCO IEP RAX TSCO TW PLL TW PLL PLD ROP PLD ROP DLPH DLPH HTZ HTZ ETE ETE WM PG HSY WM DCI VRSN PG HSY C WDC SNDK C SNDK URI MSFT LMT EPD HSIC VAL DD COL QCOM XLNX MNST LMT DD COL MNST FFIV MCK IT AKAM FFIV CCK WLL MCK NEE ASH ATVI MU AKAM TSN ETP CVX S PNW ETP INTC MKC PXD MKC PXD SO JCI ACN SO PH XEL MOLX JCI PVH OKS CCK NEE SJM CSCO S PNW ASH ATVI MU TSN IT WLL CVX SJM CSCO INTC XEL SCG D XYL RTN QCOM XLNX JBL HD ADSK KMB EMC EPD HSIC VAL NU TRIP SWKS SCG MSFT RTN TXN IP HON MXIM HD ADSK D XYL WDC TRW ADS MXIM KMB EMC CSC AVGO MWV NU TRIP SWKS BEAM URI JBL TRW HON NTAP TXN IP CSC AVGO MWV DCI VRSN NTAP BEAM ADS ACN PH MOLX PVH OKS HRL TMO HRL TMO FLS CL FLS CL PCP JNPR TEL PCP K JNPR OGE A AXP TEL K OGE A AXP VMC VMC VFC LUV WAB VFC MSM EXC FL NSC DAL FL NSC DAL BBBY HFC NOC VRTX ULTA MON RHT XRX CPA UAL JOY ULTA RHT XRX ALTR ALTR AES BX INTU EMN HSP KMI AES BX GPS EA MAS NUE VRSK MSM EXC AVP TKR VMW CLX BBBY HFC NOC VRTX MON CPA UAL JOY INTU LUV WAB AVP TKR VMW CLX EMN GPS EA MAS NUE VRSK HSP KMI CAH CAH CHD CHD NUAN NUAN IPG IPG RL PPG CI CHKP RL PPG IHS CVS CI GILD CHKP RJF GGP PXP IHS TYC LINTA UA CVS GILD RJF GGP TYC LINTA PXP SIG OMC UA SIG OMC HRS HRS BMC WINKORS CHK APH BMC BBY OII LOW BTU BMY GIS KRFT LLY FRX EFX BLL BIIB N NFLX LLTC MYL AEE FTI HOG OKE ECL SCCO LNKD Y NWL LKQ CAG FDX SRCL CPN SLB FBHS PAA RS SE WMB WRB CMCSA EQT L STJ EW STZ WAT ROK DOW URBN FDX CPN SWN WFT TIF MTD ED BMRN POM XRAY DTE TRMB DOX LNKD EQT L STJ EW EEPLNT RIG PAYX HUM RMD ECL SCCO Y SRCL GRA FAST ENR CCI WDAY HOG OKE SLB KSU CAM NBR BIIB LLTC AEE FTI SE WMB WRB COH ALB DNR MAT BLL NFLX MYL DOW URBN LKQ NI HAS NKE NBR FAST MTD ED BMRN POM XRAY DTE N ROK RIG PAYX HUM RMD NWL CAG UTX MCHP EPB HRB KSU CAM ENR CCI TRMB DOX FBHS CHRW VZ CNA CXO COH KRFT WDAY LOW BTU VMED NOV CCL FISV CPB EFX NI HAS ALB DNR MAT BAX IDXX NE SNA UTX MCHP EPB HRB NKE BBY DFS HOLX TDG VZ CNA CXO CMCSA APC ONXX ILMN LULU PNR CHRW NOV CCL FISV CPB STX FRX VMED NE SNA GIS IR LLY ILMN LULU PNR TDG APH OII BAX IDXX STX WINKORS CHK DFS HOLX IR BMY APC ONXX GRA SWN WFT STZ WAT EEPLNT OXY PAA TIF RS OXY DG DG ADP ST LUK T BEAV DO DO SYMC SYMC PX PEP PX LYB DTV BSX FNF MLM MJN OCN KBR WHR BSX FNF MLM MJN OCN HNZ VLO BRCM DLTR AMZN ALXN QGEN VLO LEN FTR KSS MMP GRMN FLEX WSM EOG GWW WLK SIAL BPL DPS SWK MO TAP LVLT BCR BHI WHR MRVL MO TAP LVLT MMP GRMN ORCL IFF ZMH BCR BHI IFF ZMH WMT MRVL FOSL WMT FOSL QEP AAPL DTV AMZN LEN FTR KSS ORCL JCP BRCM DLTR ALXN QGEN SWK MDLZ LYB FLEX WSM EOG GWW WLK SIAL BPL DPS PEP UNP JCP KBR HNZ ADP ST LUK T BEAV UNP QEP MDLZ SHW CHTR AGCO AAPL SHW CHTR AGCO CLF JBHT CLF JBHT LTD NFG KLAC IRM ISRG SYY SYK LNG FDO SYY SYK LNG TDC LTD NFG M FDO PII INGR SHLD DUK ARE EL KLAC IRM ISRG TDC SHLD DUK ARE EL M PII INGR DHR RRC DHR RRC TWX TWX SRE SRE APD WAG NVDA VAR APD COV WAG NVDA VAR COV GPC GPC COP COP BG CMS CLR CF AET BG CMS CLR SNPS CF AET SNPS JNJ MA UPS JNJ MA ACMP UPS ACMP NBL NBL CTL CTL ETR DDR SPLS FMC WPZ RSG DLR ETR DDR SPLS PETM PPL PETM PPL HES DRC APA CVI JPM DVN CELG SBH WPZ HES DRC APA CVI JPM DKS VHI FMC RSG DLR DVN CELG SBH NLSN CQP DOV AMT DKS VHI NLSN AON CCE FE TOL CQP DOV AMT AON CCE FE TOL AA AA SIRI TGT COST SBAC SIRI XEC TGT COST SBAC MDT LLL MDT LLL XEC MSI EXPE HAL ORLY ESV CE MSI AAP EXPE HAL CNH ARG ORLY ESV CE AAP CNH ARG OC OC AZO TSO AZO TSO VIAB REGN VIAB REGN LBTYA LBTYA WFM DISH WFM CA DISH CA AMP KMP AMP KMP LIFE CFN IBM LIFE GM AGN CFN IBM DE GM AGN DE GD GD PFE MRK EQIX PFE CNX MRK EQIX CNX FCX WLP RKT YHOO WGP EBAY MOS NVE FCX WLP FB RKT YHOO WGP TJX CAT MWE EBAY MOS NVE FB TJX CAT MWE MPC GOOG JEC MPC XOM MHK ABC GOOG JEC PCAR XOM MHK ABC ANSS PCAR ANSS TXT TXT F F TWC AMAT BA DELL PWR TWC AMAT BA DELL PWR NRG HPQ NRG PM HPQ WU PM WU AVT AVT GE GE CBI CBI Figure: Who are my customers (left) and suppliers (right) Green: Manufacturing, Blue: Transportation Warehousing, Red: Wholesale Retail John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 2 / 29 Outline Empirical Observations: Propagation of risk on two levels (direct and indirect) 1st-order effects (direct propagation) 2nd-order effects (systematic risk) Equilibrium Network Model Implications of the Model Conclusions and Future Directions John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 3 / 29 Empirical Observations Pricing and Risk Basics Model of share price at time t: pt = ∞ X e −(rs +δs )s ds s=0 Expected dividends ds Depends on supply chain partners (first-order). Changes may be delayed due to inattention or invisibility. Risk premium, δs Depends on multiplicity of connections to transmit risk (second-order). Reliability issues may create nonlinear effects on the risk of network position. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 4 / 29 Literature Literature 1st-order effects Industry level: Menzly & Ozbas (2007), Shahrur, Becker, & Rosenfeld (2010), Fruin, Osiol, & Wang (2012). Firm level: Hendricks & Singhal (2003), Cohen & Frazzini (2008), Atalay, Hortacsu, & Syverson (2013, working). 2nd-order effects Asset pricing: Sharpe (1964), Lintner (1965), Fama & French (1993). Network risk: Acemoglu, Carvalho, Ozdaglar, & Tahbaz-Salehi (2012), Anupindi & Akella (1993), Cachon, Randall, & Schmidt (2007), Ahern (2012), Carvalho and Gabaix (2013), Kelly, Lustig, & Nieuwerburgh (2013, working), Herskovic (2014, working). John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 5 / 29 Data Empirical Observations: Data Scope is limited to U.S. public listed firms Stock data: CRSP (monthly returns over July 2011 - June 2013) Supply chain sales data (SPLC) Compustat: SEC public filings (10% rule). Bloomberg terminal (320k units): conference call transcripts, capital market presentations, firm press releases, product catalogs, firm websites. Both are public information. SEC’s Statement of Financial Accounting Standards No. 14 (SFAS 14) “if 10% or more of the revenue of an enterprise is derived from sales to any single customer, that fact and the amount of revenue from each such customer shall be disclosed” in interim financial reports issued to shareholders John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 6 / 29 Data First-order Effects Example Relationship Customer to Supplier Calloway Golf/Coastcast (Cohen and Frazzini (2008)) Calloway misses earning forecase by half ($0.36 from $0.70). Calloway’s stock price drops 30%. Coastcast share price (50% of sales to Calloway) unchanged for one month. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 7 / 29 Data First-order Effects Example Relationship Supplier to Customer Philips/Sony/Ericsson v. Nokia Fire in Philips plant, key chip supplier for Nokia and Ericsson, in March 2000. Philips states 1-week shutdown, then revises to 6 weeks. Nokia (multi-sourcing) reacts quickly. Ericsson (single sourcing) reacts slowly, lost $2.34B, acquired by Sony. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 8 / 29 Data First-order Effects First-order Effects wijin denotes the supplier weight of j as a fraction of i’s procurement. wijout denotes the customer weight of j as a fraction of i’s sales. wijin = salesji salesji salesij salesij = PN , wijout = = PN . Procurement i Sales i k=1 saleski k=1 salesik ri,t is the return of firm i in month t. The following specification is tested: X X ri,t = α + β1 ri,t−1 + β2 wijin rj,t−1 + β3 wijout rj,t−1 j +β4 X j wijin rj,t + β5 j X wijout rj,t + i,t (1) j Hypothesis: Suppliers’ and customers’ concurrent performance relates to the firm. Supplier momentum (one-month lag) may be related to firm performance (following Cohen and Frazzini (2008)). John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 9 / 29 Data First-order Effects First-order Effects Results Table: Fama-Macbeth Regression of Concurrent Returns and Momentum. α Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) Ave. Coef (T-Stat) -0.001 (-0.96) 0.009*** (10.38) 0.009*** (10.53) 0.008*** (11.09) 0.008*** (10.92) 0.003*** (3.61) -0.002** (-2.26) 0.004*** (4.51) -0.002* (-1.92) -0.001* (-1.80) ri,t−1 -0.088*** (-11.06) -0.090*** (-9.08) -0.047*** (-6.96) P in j wij rj,t−1 0.036** (2.17) 0.057*** (2.96) P out j wij rj,t−1 0.024 (0.95) 0.004 (0.09) P in j wij rj,t 0.399*** (20.90) P out j wij rj,t 0.755*** (3.12) 0.022** (1.83) -0.040 (-0.66) 0.619*** (37.25) 0.992*** (4.54) 0.018* (1.57) 0.625*** (36.44) 0.001 (0.0274) 0.393*** (22.48) 1.001*** (4.51) 0.744*** (3.20) *p-value<10%, **p-value<5%, ***p-value<1% Controls: MKT, SMB, HML, MOM, cross-firm effect, industry effect. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 10 / 29 Data Second-order Effects Second-order Effects Assumption Acemoglu, Carvalho, Ozdaglar, & Tahbaz-Salehi (2012) (and the extension later) finds that microeconomic idiosyncratic shocks lead to aggregate flucturations, which means A firm’s systematic risk is formed from the aggregation of idiosyncratic shocks. Effects of connections may be nonlinear due to interactions - risk diversification or aggregation? Firm level shocks may be exogenously correlated due to geographical proximity and sector proximity. A manufacturer (e.g., Nokia) may have diversification incentives to add an independent supplier to increase reliability (reduce systematic risk exposure with greater centrality). A distributor (e.g., a beverage distributor) may have concentration incentives to add similar suppliers (e.g., French wineries) to build on existing capabilities (increase systematic risk exposure with greater centrality). John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 11 / 29 Data Second-order Effects Second-Order Results: Manufacturing Table: Factor Sensitivities by Eigenvector Centrality for Manufacturing Firms. N3 Portfolio 1(High) 2 3 4 5(Low) High-Low α (%) 0.235 (1.50) 0.114 (0.49) 0.295* (1.78) 0.277 (1.34) 0.328 (1.33) 0.482* (1.86) 0.356 (0.89) 0.571 (1.36) 0.507 (1.55) 0.934* (1.95) -0.272* (-1.72) -0.820* (-1.96) Factor Loadings Rmt − Rft SMB HML 0.888*** (15.47) 0.894*** -0.347* 0.018 (12.23) (-2.07) (0.119) 0.773*** (13.79) 0.938*** -0.184 -0.453*** (14.28) (-1.22) (-3.29) 1.060*** (17.60) 0.953*** 0.363* -0.005 (11.63) (1.93) (-0.03) 1.256*** (12.97) 1.087*** 0.446 0.130 (8.22) (1.47) (0.47) 1.410*** (11.96) 1.157*** 0.780** -0.257 (7.63) (2.24) (-0.80) -0.522 (-3.92) -0.263 -1.127** 0.275 (-1.28) (-2.40) (0.64) *p-value¡10%, **p-value¡5%, ***p-value¡1% MOM 0.084 (1.025) Adj. R 2 (%) 90.85 90.01 88.74 -0.061 (-0.83) 93.77 -0.008 (-0.09) 93.04 -0.142 (-0.96) 87.82 92.78 87.45 85.54 -0.132 (-0.78) 87.53 0.216 (0.94) Controls: MKT, SMB, HML, MOM, cross-industry concentration. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 12 / 29 Data Second-order Effects Second-Order Results: Logistics Table: Factor Sensitivities by Eigenvector Centrality Centrality for Logistics Firms. N4 Portfolio 1(High) 2 3 4 5(Low) High-Low Alpha(%) 1.314*** (3.26) 1.428*** (3.44) 0.894*** (3.78) 0.916*** (2.41) 0.812** (2.23) 0.801** (3.36) 0.708** (2.50) 0.669** (2.14) 0.759 (1.44) 0.485 (0.84) 0.556 (1.53) 0.975* (1.93) Factor Loadings Rmt − Rft SMB HML 0.747*** (7.62) 0.768*** 0.006 -0.589 (5.85) (0.02) (-2.14) 0.671*** (11.67) 0.976*** 0.034 -0.502 (8.13) (0.13) (-1.99) 0.964*** (10.89) 0.758*** -0.140 -0.152 (10.03) (-0.81) (-0.96) 0.857*** (12.40) 0.916*** -0.171 -0.190 (9.26) (-0.75) (-0.92) 0.776*** (6.03) 0.942*** -0.548 0.141 (5.17) (-1.31) (0.37) -0.029 (-0.20) -0.175 0.553 -0.730 (-0.90) (1.24) (-1.69) *p-value¡10%, **p-value¡5%, ***p-value¡1% MOM Adj. R 2 (%) 84.93 0.024 (-0.16) 86.43 0.031 (0.23) 72.32 0.164 (1.93) 83.75 0.019 (0.17) 85.49 70.41 83.05 86.41 69.60 0.048 (0.23) 67.70 -0.024 (-0.11) Controls: MKT, SMB, HML, MOM, cross-industry concentration. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 13 / 29 Data Second-order Effects Second-Order Results: Mining, Utilities, and Construction Table: Factor Sensitivities by Eigenvector Centrality for NAICS 2 Industries. N2 Portfolio 1(High) 2 3 4 5(Low) High-Low α (%) -1.153* (-1.74) -1.179 (-1.52) -0.897 (-1.25) -1.023 (-1.21) -0.346 (-0.62) -0.680 (-1.09) -0.374 (-0.58) -0.598 (-0.83) -0.479 (-0.72) -0.626 (-0.82) -0.674* (-1.95) -0.553 (-1.51) Rmt − Rft 1.399*** (9.09) 1.458*** (5.80) 1.512*** (9.06) 1.583*** (5.76) 0.762*** (5.93) 0.935*** (4.63) 1.129*** (7.58) 1.213*** (5.20) 1.339*** (8.74) 1.456*** (5.90) 0.060 (1.25) 0.002 (0.02) Factor Loadings SMB HML -0.091 (-0.15) -0.330 (-0.68) MOM 0.114 (0.45) Adj. R 2 (%) 79.54 76.83 79.42 -0.329 (-0.48) 0.092 (0.17) -0.103 (-0.37) 76.28 61.90 -0.458 (-0.92) 0.262 (0.67) 0.155 (0.76) 59.96 72.88 -0.071 (-0.12) -0.376 (-0.84) 0.261 (1.11) 71.72 78.22 -0.201 (-0.33) -0.215 (-0.45) 0.221 (0.89) 0.110 (0.51) -0.115 (-0.68) -0.107 (-1.20) 75.94 *p-value¡10%, **p-value¡5%, ***p-value¡1% John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 14 / 29 Data Second-order Effects Centrality Implications to Different Industries Both shock correlation and network topology matter for systematic risk. Methodology: Expected returns should be explained by all systematic risk factors. Split firms into quintiles based on centrality measures. If ∆α 6= 0 for two extreme quintile portfolios, supply chain network leads to ”anomalies” in systematic risk. Positions in the supply chain affects a firm’s exposure to the systematic risk besides the network topology. Upstream firms in manufacturing have diversification incentive to form supplier connections to operationally hedge risk. Downstream firms in logistics have concentration incentive to form supplier connections to leverage economy of scale thus aggregate risk. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 15 / 29 Equilibrium Network Model Simple Investment Model Suppose an economy with 2 regions (A and B) and 3 potential future states below with equal probability (Prob (S = Si ) = 13 , ∀i ∈ {1, 2, 3}): S1 : both A and B function; S2 : A cannot produce and B can; S3 : B cannot produce and A can. Next, suppose 4 firms: 3 manufacturers and 1 distributor. Manufacturers: limited capacity and payoff of 1 as long as one input region functions. Sources: Firm 1 only from region A Firm 2 only from region B Firm 3 from both Firm 4 is the distributor and connects to both A and B with a fixed cost of 1 in all states. Payoffs: Π1 = {1, 0, 1}, Π2 = {1, 1, 0}, Π3 = {1, 1, 1}, Π4 = {1, 0, 0}. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 16 / 29 Equilibrium Network Model Investment Model Solution Let Ω denote the covariance matrix for the firms’ payoffs. 1 − 16 0 16 3 1 −1 0 16 6 3 Ω= 0 0 0 0 1 1 0 13 6 6 Suppose we have a representative mean-variance investor, and let µ = [µ1 , µ2 , µ3 , µ4 ] denote firms expected return. Then for any feasible returns µ̃ the investor targets, the investor find the portfolio weights w = [w1 , w2 , w3 , w4 ] by solving 0 0 0 min{w Ωw |w µ = µ̃; w 1 = 10 } w , John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 17 / 29 Equilibrium Network Model Simple Investment Results The result of the equilibrium of investment is: 1 1 µ1 6 w1 + 6 w4 1 µ2 1 = 1 6 w1 + 6 w4 + λ 2 µ3 λ1 λ1 0 1 1 µ4 w + w 3 1 3 4 Therefore, µ3 < µ1 = µ2 < µ4 i.e. the manufacturers have lower risk than the distributor, and the dual sourcing manufacturer is less risky than the single sourcing manufacturer. Questions: Does this result generalize to a broader classs of networks and what are other empirical implications? Are the output representations consistent with an equilibrium model of production? John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 18 / 29 Equilibrium Network Model Equilibrium Network Model and Relationship to Literature Previous literature focus on the sector level only. Lucas (1977) argues that microeconomic shocks would average out at the aggregated level proportional to √1n . Acemoglu et al. (2012) suggests Lucas (1977) only holds under symmetric network structure, and microeconomic shocks may lead to aggregated fluctuations in asymmetric networks. The change in the density of firm level connections is not captured. We build a supply chain network model using two-level nested production function capturing both the firm-level and sector-level connections. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 19 / 29 Equilibrium Network Model Model Setup An extension of Acemoglu et al. 2012 n industry sectors (S1 , S2 , ..., and Sn ). Firms in the same sector have the same Cobb-Douglas CRS production function to produce perfectly substitutable products. Supply chain relationships are established ex-ante. xijkl : output from firm l in sector j that inputs to firm k in sector i. P xi = k∈Si xik : output from firm k in sector i. P P xij = k∈Si l∈Sj xijkl : the production from sector j to sector i. P xi = k∈Si xik : sector i’s total production. k A unit P labor kallocating Pn to to each firm (li ) in each sector (li ), i.e. li = k∈Si li and i=1 li = 1. P Consumption from by firm k in sector i is cik , and ci = k∈Si cik . Total consumption / GDP / labor wage is h. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 20 / 29 Equilibrium Network Model Competitive Equilibrium A competitive equilibrium of economy We define a competitive equilibrium of economy with nPsectors consisting of prices (pi , i ∈ {1, ..., n}), wage h, consumption bundle ci = k∈Si cik , ∀i, k ∈ Si , and quantities lik , xik , xijkl , ∀i, j, k, l such that 1 2 3 the representative consumer maximizes her utility; the firms in each sector maximizes their profits (0 in expectation); the labor and good markets clear at both levels, i.e. for any firm k in any sector i, and for any sector i, xik = cik + n X X xjilk , j=1 l∈Sj xi = c i + n X j=1 John Birge and Jing Wu (Chicago Booth) X lik = li k∈Si xji , n X li = 1 i=1 Equilibrium in Supply Chain Networks June 20, 2016 21 / 29 Equilibrium Network Model Household and Firm Problems The customer has Cobb-Douglas preferences over distinct goods from n sectors subject to budget constraint, that is 1 max u (c1 , c2 , ..., cn ) = AΠni=1 (ci ) n , s.t. n X pi ci ≤ h i=1 Firm problem solves the following maximization problem maxΠki = pi xik − hlik − n X j=1 xik = zik X xijkl l∈Sj (1−α)wij n X Y α xijkl lik j=1 John Birge and Jing Wu (Chicago Booth) pj l∈Sj Equilibrium in Supply Chain Networks June 20, 2016 22 / 29 Equilibrium Network Model From Firm Connections to Sector Connections Since firms face the same input prices and own the same production technology, they will choose the same proportions of inputs: X xijkl = γik xij , lik = γik li l∈Sj P where γik = l∈Sj xijkl xij = lik li is the firm’s sector share. Firm-level networks determine the shape of the sector shock distribution. The Origin of Sector Shock α Qn (1−α)wij In sector i’s output, i.e. xi = zi (li ) , the sector productivity j=1 (xij ) shock is a sum of firm level shocks, weighted by each firm’s sector share. X zi = γik zik k∈Si John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 23 / 29 Equilibrium Network Model From Firm Connetions to Sector Connetions Firm-level connections affect the sector shock through the distribution of the firm’s sector share γik . 0 Define the influencePvector as v = n vi = Pnpi xpi i xi thus i=1 vi = 1. α 0 n1 −1 [I − (1 − α) W ] satisfying i=1 Supply Chain Network Systematic Risk The aggregate output is a influence vector weighted sum of sector-specific productivity shocks below. 0 y = lnh = v P k k where is a column vector with i = lnzi = ln k∈Si zi γi . The volatility of the aggregate output (the systematic risk) is h 0 i Var [y ] = Var v John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 24 / 29 Equilibrium Network Model Sparse v.s. Dense Supply Chain Networks Sector A Sector A Sector B Sector B The firm’s sector share γik in the left case would have higher variance on the distribution than the right case. Similar to Proposition 4 inh Acemoglu et al. (2015), the expected total output i 0 E [y ] decreases when Var v increases, i.e. 1. Supply Chain Network and Sector Performance For concave production functions, a sparse firm-level supply chain network results in less total sector output than a dense network. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 25 / 29 Equilibrium Network Model Simulation Step 1 (Relationship-formation): Each firm chooses a set of suppliers. Ex-ante the market is perfectly competitive. Step 2 (Input-acquisition): Each firm draws i.i.d. production shock. Input quantity depends on the supplier actual production. The dense network has a low sector weight variance (std 0.0001 v.s. 0.0023). 2. Supply Chain Network and Firm Volatility A sparse network results in more volatile firm production than a dense network. 1400 1200 1200 1000 1000 800 800 600 600 400 400 200 200 0 8.8 9 9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 −3 0 0 0.005 0.01 x 10 0.015 0.02 0.025 0.03 0.035 Figure: Sector Weight γik Distribution (Left: 80% of Suppliers, Right: 2% of Suppliers). John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 26 / 29 Equilibrium Network Model Simulation (cont.) Both cases exhibit sizable and systematic deviations from the normal distribution (2%: heavy left tail, 80%: heavy right tail). Only the sector weight with modest connection density is normal distributed. Sufficient Statistics for Firm Production Variation With firm-level supply chain connection variation, there is no guarantee that the firm-level production is normally distributed. −3 10.8 QQ Plot of Sample Data versus Standard Normal x 10 QQ Plot of Sample Data versus Standard Normal 0.025 10.6 0.02 Quantiles of Input Sample Quantiles of Input Sample 10.4 10.2 10 9.8 9.6 9.4 0.015 0.01 0.005 9.2 9 −4 −3 −2 −1 0 1 Standard Normal Quantiles 2 3 4 0 −4 −3 −2 −1 0 1 Standard Normal Quantiles 2 3 4 Figure: Q-Q Plot of the Sector Weight γik Distribution (Left: 80% of Suppliers, Right: 2% of Suppliers). John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 27 / 29 Equilibrium Network Model More Concentrated Economic Activities during Crisis. Core: most (eigenvector) central firms; Periphery: least central firms. Force-directed layout algorithm (Fruchterman and Reingold 1991). Left: network in July 2007; Right: network in June 2009. Economic activities for June 2009 supply chain network are more concentrated than July 2007. John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 28 / 29 Conclusions & Future Directions Conclusions and Future Directions Evidence of concurrent supplier and customer effects plus supplier momentum effects on returns. Investors’ limited attention to supplier firms relative to customer firms. Gradual diffusion of supply chain information downstream as opposed to upstream. Evidence of decreasing returns to centrality in manufacturing and increasing returns to centrality in logistics. Supply chain structure is an ex-ante determined and ex-post identifiable source for systematic risk. Upper-stream utility, mining, and construction firms behave similarly as manufacturing. Equilibrium model of firms connecting across sectors Natural hedging decisions from manufacturers Lower volatility effect for manufacturers implies conditions for increasing conections to cause lower risk for upstream and higher risk for downstream Future Directions Additional empirical tests (including default propagation) Incorporate of investment into the model formulation John Birge and Jing Wu (Chicago Booth) Equilibrium in Supply Chain Networks June 20, 2016 29 / 29