Network Externality and Subsidy Structure in Two-Sided Markets:

Transcription

1 Network Externality and Subsidy Structure in Two-Sided Markets: Evidence From Electric Vehicle Incentives Katalin Springel Resources for the Future October 20, 2017

2 Introduction Motivation Environmental impact of transportation Governments are using a variety of incentives Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 1 / 18

3 Introduction Market Sales Shares of EVs (2016) Percentage (%) China Denmark France Germany Japan Netherlands Norway Sweden UK US Source: International Energy Agency (2017) Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 1 / 18

4 Introduction Two-Sided Market for EVs CarManufacturers actur Buyers/Drivers Feedback Efects ChargingStations Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 1 / 18

5 Introduction Motivation Environmental impact of transportation Governments are using a variety of incentives Research Question What is the most effective subsidy structure in a two-sided market with network externalities? Specifically, is it preferable to directly subsidize buyers or stations? Focusing on the Norwegian car market Preview of results Station subsidies are more than twice as effective as price subsidies However, the relation inverts with increased spending Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 1 / 18

6 Policy Background Electric vehicle supply equipment (EVSE) incentives Differentiated by rate of charging (normal vs. fast) Both at national level (since 2009) and in selected counties Purchasing price subsidies to consumers Registration tax exemption (since 1990) Reduced/No annual license fee (since 1996) VAT exemption (since 2001) Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 2 / 18

7 Policy Background Examples of Tax Savings (2015) Vehicle Model MSRP Total Tax Tesla Model S Luxury/Sports Cars $78,343 $49 BMW 7-Serie 750i $77,653 $94,430 Compact Cars Nissan Leaf $24,973 $49 Volkswagen Golf $22,270 $13,151 Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 3 / 18

8 Data Characteristics Time period: Geographic level: County Vehicle registry Newly registered cars of all fuel types Vehicle specification: Maker/model/trim Includes all technical characteristics and MSRP Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 4 / 18

9 Vehicle Data Cumulative All-Electric Vehicle Sales in Norway Cumulative Number of AEVs Sold /2010 1/2011 1/2012 1/2013 1/2014 1/2015 1/2016 AEV PHEV Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 4 / 18

10 Data Characteristics Time period: Geographic level: County Vehicle registry Newly registered cars of all fuel types Vehicle specification: Maker/model/trim Includes all technical characteristics and MSRP Station data Number of charging stations/outlets Station characteristics (location, connector type, owner) Descriptives Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 4 / 18

11 Station Data All-Electric Vehicle Sales and Number of Stations in Norway Number of Charging Points Cumulative AEV Sales 0 1/2010 1/2011 1/2012 1/2013 1/2014 1/2015 1/ Number of Charging Points AEV Sales Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 5 / 18

12 Station Data Norwegian Station Network in 2009 Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 6 / 18

13 Station Data Norwegian Station Network in 2015 Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 7 / 18

14 Descriptive Analysis To explore the relation between EV subsidies and EV purchases I analyze vehicle sales of all fuel types as a function of EV incentives Macroeconomics variables Rich set of time and county-by-model fixed effects Main result of analysis $1,200/car increase in tax exemption leads to a 3.09% increase in EV sales $1,200/station increase in station subsidies increases EV sales by 8.42% Full Table Robustness Checks Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 8 / 18

15 The Setup Simultaneous decision-making Consumers make vehicle purchase decision Stations make entry decision Consumers choose their demand for charging and stations provide charging Two obstacles this analysis needs to address Price correlates with unobserved car attributes Station entry and EV sales are determined simultaneously Consideration of dynamics Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 9 / 18

16 Vehicle Demand Model Consumers decide whether to buy a new vehicle or not The indirect utility of a consumer is defined as a function of Consumer s income Purchasing price Charging network Observed and unobserved attributes Idiosyncratic shocks Random coefficients discrete choice model Builds on Berry, Levinsohn & Pakes (1995) Flexible substitution patterns Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 10 / 18

17 Vehicle Demand Model Solution to the two identification problems 1. Instruments for the price Based on observed exogenous car attributes Instrument validity is violated if observed characteristics are chosen after unobserved attributes are known 2. Instruments for the network effects EVSE incentives Instrument validity is violated if policymakers change incentives as unobserved EV demand changes Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 11 / 18

18 Station Entry Model Station entry depends on sunk entry costs and the discounted stream of profits Following Gandal et al. (2000) and Bresnahan & Reiss (1991) Station entry is modeled as a function of Cumulative EV base Station subsidies County and time-specific factors Idiosyncratic shocks Instrument for feedback effects Gas station density Instrument validity is violated if some unobserved time-invariant factors correlate with both the IV and the charging network Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 12 / 18

19 Results Feedback effects are positive and significant Charging network influences buyers vehicle choice Heterogeneous valuation of the network Station entry is positively responsive to the EV base EV models can act as complements Only when there are strong feedback effects Expected to weaken and disappear over time Structural Estimates Sample Elasticites Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 13 / 18

20 Policy Counterfactuals Consider alternative incentive structures: 1. EV price discounts (e.g. tax exemption) 2. Station cost discounts (e.g. direct one-time subsidies) Jointly determine equilibrium number of stations and vehicle sales under new incentive structures Set discounts so that there is revenue equivalence between incentive schemes Compare how changes in the two subsidies affect total EV sales, given a fixed amount of resources Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 14 / 18

21 Policy Counterfactuals (Average Impact) Status quo Current Subsidies Counterfactuals Only Price Subsidy Only Station Subsidy [1] [2] [3] ΔTotal EV Purchases 18,005 16, ΔTotal Stations ΔTotal Government Spending (Million NOK) 4,552 4, ΔEV Purchases / ΔGovernment Spending Normal EVSE incentives Y N Y Car Purchase Incentives Y Y N Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 15 / 18

22 Policy Counterfactuals (Marginal Impact) 4,000 Change in EV Purchases 3,000 2,000 1, ,000 Change in Government Spending (million NOK) EV Price Subsidy Station Subsidy Further Analysis Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 16 / 18

23 Policy Counterfactuals (Marginal Impact) The Impact of Different Subsidy Allocations Change in EV Sales Change in Government Spending Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 17 / 18

24 Conclusion What do we learn? Network effects play a key role in the EV market Feedback effects result in complementarities between EVs The feedback loops further amplify the effect of subsidies Heterogeneity in consumer valuation is similarly important Provides more realistic substitution patterns Affects the impact price subsidies have on demand for EVs Counterfactual policy analysis shows that Effectiveness of subsidies is an open empirical question Katalin Springel (RFF) Network Externality and Subsidy Structure in Two-Sided Markets 18 / 18

25 Thank you! Comments and questions appreciated Katalin Springel

26 Descriptive Statistics Year Mean No. of Models Sales Stations Price HP/Wt Consumption EV Length Transmission ,763 2, , ,976 3, , ,451 3, , ,383 4, , ,592 6, , ,614 7, , Note: The entry in each cell of the last six columns is the vehicle sales weighted mean. Back

27 Reduced-Form Analysis log R jct = α + E V β E V jct + G Y µ G Y ct + ϑ jc + ϑ t + ϵ jct Notation: Unit of obs: model j in county c at time period t R = number of newly registered vehicle models V = set of available AEV incentives Y = macroeconomics controls ϑ jc = county-by-model FE ϑ t = time FE

28 Results from the Reduced-Form Analysis log(no. of Registered Cars) [1] [2] [3] Registration Tax Exemption (10,000 NOK) *** *** *** ( ) ( ) ( ) VAT Exemption (10,000 NOK) (0.0146) (0.0146) (0.0147) EVSE Normal (10,000 NOK) * * (0.0171) (0.0156) (0.0152) EVSE Normal EV 0.168*** *** *** (0.0263) (0.0275) (0.0276) EVSE Fast (10,000 NOK) ( ) ( ) ( ) EVSE Fast EV ( ) ( ) ( ) Observations 191, , ,616 Adj. R-squared Model County and Time Fixed Effects Y Y Y Cluster on Model and County Y Y Y Local Incentives N Y Y Macroeconomic Controls N N Y Back

29 Robustness Checks for the Reduced-Form Analysis Potential concern: Identifying assumption is violated due to confounding factors. Robustness checks Interaction of policy terms with a hybrid vehicle dummy

30 Robustness Checks for the Reduced-Form Analysis log(no. of Registered Cars) EVSE Normal (0.0119) EVSE Normal Hybrid (0.0465) EVSE Fast (0.0007) EVSE Fast Hybrid (0.0031) Observations 191,616 Adj. R-squared 0.39 Model-County and Time Fixed Effects Y Cluster on Model and County Y Local and Tax Incentives Y Macroeconomic Controls Y

31 Robustness Checks for the Reduced-Form Analysis Potential concern: Identifying assumption is violated due to confounding factors. Robustness checks Interaction of policy terms with a hybrid vehicle dummy Randomized assignment of station subsidies

32 Robustness Checks for the Reduced-Form Analysis log(no. of Registered Cars) Placebo EVSE Normal (0.0019) Placebo EVSE Normal EV (0.0171) Placebo EVSE Fast (0.0003) Placebo EVSE Fast EV (0.0013) Observations 191,616 Adj. R-squared 0.40 Model-County and Time Fixed Effects Y Cluster on Model and County Y Local and Tax Incentives Y Macroeconomic Controls Y

33 Robustness Checks for the Reduced-Form Analysis Potential concern: Identifying assumption is violated due to confounding factors. Robustness checks Interaction of policy terms with a hybrid vehicle dummy Randomized assignment of station subsidies Lead and lagged versions of station subsidies

34 Robustness Checks for the Reduced-Form Analysis log(no. of Registered Cars) EVSE Normal EV ** (0.0540) Lead EVSE Normal EV (0.0322) Lagged EVSE Normal EV (0.1056) Observations 62,758 Adj. R-squared 0.45 Model-County and Time Fixed Effects Y Cluster on Model and County Y Local and Tax Incentives Y Macroeconomic Controls Y Back

35 Structural Estimation Results Vehicle Demand Station Entry Variable Parameter Estimate Standard Error log(income - Price) Station Network (Mean) Station Network (Std. Deviation) Variable Parameter Standard Estimate Error log(ev base) EVSE normal (10,000 NOK) EVSE fast (10,000 NOK) Trend Back

36 Sample of Own- and Cross-Price Semi-Elasticities AEV Make and Model i3 C-Zero i-miev Leaf Ion E-Up! No Feedback Effects BMW i Citroen C-Zero Mitsubishi i-miev Nissan Leaf Peugeot Ion Volkswagen E-Up! With Feedback Effects BMW i Citroen C-Zero Mitsubishi i-miev Nissan Leaf Peugeot Ion Volkswagen E-Up! Each cell entry, where i denotes rows and j denotes columns, shows the percentage change in market share of model i with a 10, 000 NOK change in price of model j. Back

37 Policy Counterfactuals (Marginal Impact) 1,000 Change in Number of Stations ,000 Change in Government Spending (million NOK) EV Price Subsidy Station Subsidy

38 Policy Counterfactuals (Marginal Impact) Percent Decrease in EV Price Percent Increase in Station Subsidy ,000 Change in Government Spending (million NOK) 0 EV Price Subsidy Station Subsidy Back

39 Real-time Usage - U.S. Public Charging Network (1) Percentage (%) Hours

40 Real-time Usage - U.S. Public Charging Network (2) Percentage (%) Sun Mon Tue Wed Thu Fri Sat In Use Out of Service Available

41 Theoretical Framework Two-sided markets model Platform Facilitates the interaction between the two sides Sets prices In case of AEVs Drivers one side Charging stations other side Car manufacturers/aev technology platform

42 Theoretical Framework Key features Acts of users on one side affect users on the other side Non-neutrality of price allocation Pure Membership Externality Users on each side bring external benefits to the other simply by participating

43 Key Assumptions Two-sided market Pure membership externality & pricing Linear network effects One-dimensional heterogeneity of users

44 Baseline Model (1) I {B, S} and J = I # of agents on side I: N I P B B B Platform P S Membership value: BI i Membership fee: P I Marginal cost: C I Interaction value: b i I = b I Buyer (B) B S Station (S)

45 Baseline Model (1) I {B, S} and J = I # of agents on side I: N I P B B B Platform P S Membership value: BI i Membership fee: P I Marginal cost: C I Interaction value: b i I = b I Buyer (B) B S Station (S)

46 Baseline Model (1) I {B, S} and J = I # of agents on side I: N I P B B B Platform P S Membership value: BI i Membership fee: P I Marginal cost: C I Interaction value: b i I = b I Buyer (B) B S Station (S)

47 Baseline Model (1) I {B, S} and J = I # of agents on side I: N I P B B B Platform P S Membership value: BI i Membership fee: P I Marginal cost: C I Interaction value: b i I = b I Buyer (B) B S Station (S)

48 Baseline Model (1) I {B, S} and J = I # of agents on side I: N I P B B B Platform P S Membership value: BI i Membership fee: P I Marginal cost: C I Interaction value: b i I = b I Buyer (B) B S Station (S)

49 Baseline Model (1) I {B, S} and J = I # of agents on side I: N I P B B B Platform P S Membership value: BI i Membership fee: P I Marginal cost: C I Interaction value: b i I = b I Buyer (B) B S Station (S)

50 Baseline Model (2) Utility function of an agent i on side I U i I = b I N }{{ J } + BI i }{{} P }{{} I net interaction benefit membership value membership fee Source of user heterogeneity: B i I Membership value B i I is drawn from cdf F I before membership decision has been made

51 Baseline Model (3) Demand functions can be defined as N I = P r(b I N J + B i I P I ) D I (N J, P I ) Continuum of users on each side, with mass normalized to 1 N I represent the fraction of potential users that join

52 Baseline Model (4) Monopolist platform s profit function π platform = (P B C B )N B }{{} + P S N }{{ S } profit from Buyers/Drivers profit from Sellers/Stations Platform sets the prices and end-users on both sides are price-takers

53 Government Incentives (1) Comparison of subsidy schemes in the model: Subsidies to buyers only (σ B, 0) Subsidies to sellers only (0, σ S ) Additional assumption: Incentives are government revenue equivalent T = σ B N B(σ B, 0) = σ S N S(0, σ S )

54 Government Incentives (2) New utility function of an agent i on side I U i I = b I N }{{ J } + BI i }{{} net interaction benefit membership value P I }{{} membership fee + σ I }{{} subsidy Monopolist platform s profit function stays the same

55 Government Incentives (3) Solving for equilibrium with subsidies: 1. Solve for N B = ˆD B (P B, P S, σ B, σ S ) and N S = ˆD S (P B, P S, σ B, σ S ) 2. Substitute into platform s profit function 3. Solve for (P B (σ B, σ S ), P S (σ B, σ S )) 4. Get equilibrium participation rates: N B (P B, P S ), N S (P B, P S )

56 Government Incentives (4) Solving for equilibrium with subsidies: 5. Solve for σ B and σ S subject to σ B N B(P B(σ B, 0), P S(σ B, 0)) = σ S N S(P B(0, σ S ), P S(0, σ S )) 6. For all pairs (σ B, σ S ) obtained in (5), plot N B (σ B, 0) on the x-axis and N B (0, σ S) on the y-axis. 7. Neutrality of subsidies would imply all points lie on the 45 line.

57 Government Incentives (5) Assume linear demand fn: B i B iid U[µ B, ϕ B ], B i S iid U[µ S, ϕ S ] Assume nonnegative subsidies σ B, σ S Set arbitrary values for parameters: Membership benefits µ B, ϕ B, µ S, ϕ S Net interaction benefits b B, b S Marginal cost C B

58 Government Incentives (5) Examples (for arbitrary parameter values): NB ( 0, σs) NB ( σb, 0)

59 Government Incentives (5) Examples (for arbitrary parameter values): NB ( 0, σs) NB ( σb, 0)

60 Government Incentives (5) Examples (for arbitrary parameter values): NB ( 0, σs) NB ( σb, 0)

61 Government Incentives (6) Non-neutrality in allocation of subsidies Since the structure of subsidies between the two sides of the market matters for consumers vehicle purchase decision The question is which incentive is more effective in promoting EV adoption? I construct a structural model which encompasses both sides of the market to estimate the impact of the two policies Back

62 Motivation - Policy Goal With more research and incentives, we can break our dependence on oil with biofuels, and become the first country to have a million electric vehicles on the road by President Barack Obama, 2011 State of the Union