Consumer governance in electricity markets

Energy Economics 56 (2016) 326–337

Contents lists available at ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneco

Consumer governance in electricity markets夽 Toby Daglish Victoria University of Wellington, New Zealand

A R T I C L E

I N F O

Article history: Received 3 May 2015 Received in revised form 30 December 2015 Accepted 18 March 2016 Available online 6 April 2016 JEL classification: Q41 C35 M31 M38

A B S T R A C T This paper examines switching decisions by households in the MainPower distribution area of New Zealand. The paper measures the extent to which customers switch retailers following the release of information about directors’ bonuses, marketing surrounding firm ownership, and work by the New Zealand Electricity Authority to promote transparency of the switching process. We document strong customer inertia, which, for some consumers, has reduced following the Electricity Authority’s interventions. Customer movements following information releases and marketing campaigns are modest, suggesting that prices and inertia may be the most important drivers of customer migration. © 2016 Elsevier B.V. All rights reserved.

Keywords: Electricity retail Switching behaviour Customer inertia Price sensitivity

1. Introduction Electricity is fundamentally a homogeneous product. Retail consumers should, in principle, be indifferent between who they buy their electricity from, barring price considerations. This paper uses individual meter level switching data to explore the extent to which this hypothesis reflects reality, and, in particular, whether customers are also motivated by the ownership/management behaviour of retailers. Retail customers play an important role in the risk-management strategies of gentailer (vertically integrated generator/retailer) electricity companies. Firms who have retail market shares equal to their production shares are less inclined to exert market power in whole-

夽 The author thanks Yi˘git Sa˘glam, Oliver Robertson, Dayle Parris, Jean-Pierre Florens, Beverly Brereton, Andrew Kerr, Harold Cuffe, Phil Bishop, James Bushnell, Frank Wolak, Gabriel Fiuza de Bragança, participants at the ATE workshop 2013, TIGER forum 2014, along with staff of Meridian Energy, IPEA, and the New Zealand Electricity Authority for their feedback at seminars. The author thanks MainPower, especially Dayle Parris and Stephen Lewis (from PowerSwitch), for providing customer switching data, Stephen Burrough for providing assistance with identifying price change dates, and Vaibhav Kawale (from Consumer NZ) for data on customer satisfaction. Geographic Information Systems support was provided by Richard Law. Funding for this work was provided by the New Zealand Institute for the Study of Competition and Regulation, and much of the research was undertaken while the author was employed there. E-mail address: [email protected] (T. Daglish).

http://dx.doi.org/10.1016/j.eneco.2016.03.018 0140-9883/© 2016 Elsevier B.V. All rights reserved.

sale markets.1 They are also, given the fixed prices negotiated with customers, relatively immune to fluctuations in wholesale prices (Bushnell et al., 2008). Risk management may encourage gentailers to target retail market shares in line with their wholesale market shares. More aggressive firms may choose to take a position that leaves them as net buyers or sellers in the wholesale market. In either case, however, firms will frequently have a desired level of retail market penetration. A regulator concerned with market power in the wholesale market may desire firms’ market shares to align with their productive capacity. However, the regulator may also be concerned about retail pricing, and so may encourage customers to move retailers. Giulietti et al. (2010a) show that, in the English/Welsh market, there is a weak link between wholesale prices and retail prices, suggesting that market power in the retail market is an important determinant of consumer prices. Active customer switching encourages firms to compete on prices, but also in the service they provide, hopefully leading to higher customer satisfaction in the market. Satisfied customers not moving may not be a cause for regulatory concern. However, the presence of customers with low satisfaction, paying higher

1 Hortaçsu and Puller (2008) argue that firms will not exert market power in the wholesale electricity market when their productive capacity is perfectly hedged.

T. Daglish / Energy Economics 56 (2016) 326–337

costs than they could obtain from a competing retailer, may suggest that customer welfare could be improved by informing customers of their ability to move retailers. The behaviour of customer switching in electricity markets is thus of interest to both firms targeting a particular market share, and to regulators who may be targeting another particular set of market shares, and hoping to achieve a healthy level of “churn” among customers. Brennan (2007) reports that most jurisdictions where electricity customers can choose a retailer, with the exception of Britain, have exhibited very low switching rates, as customers seem reluctant to move from their initially assigned retailer (the region’s “incumbent” retailer). He notes that New Zealand has the second highest (after Britain) switching rate, but that this still leaves 75% of customers having not switched (at time of writing).2 Defeuilley (2009) attributes low switching rates and suboptimal behaviour of households to behavioural biases encouraging customers to stick with the “status quo”, and risk aversion. He also comments that in many cases retail offerings have been less innovative than would be expected, and customers may not have been offered something “new” by incoming retailers. Defeuilley’s view is consistent with Giulietti et al. (2005)’s survey evidence that search costs are perceived to be high for many customers. In contrast, Littlechild (2009) remarks that most deregulated markets are seeing growing switching behaviour over time. Economically, switching propensities are important for retailers: Giulietti et al. (2010b) document that since privatisation, in the UK market, prices for incumbents have not converged with entrant prices, suggesting that an incumbent retailer can exploit inertia in customer switching decisions.3 Wilson and Waddams Price (2010) show that although most customers who switch retailers are motivated by costs, a fifth of customers lose economic surplus by switching. Waddams Price and Zhu (2016) note that UK retailers (prior to 2009, when the practice was banned) priced higher in their “home” jurisdictions than they did when competing in other regions (presumably to exploit inertia). We hypothesise that retail customers are motivated by several factors in making a decision to switch retailers. First, we assume that customers are motivated by price concerns. If a competitor offers a lower price, customers will be inclined to switch retailer. This is mitigated by search costs, and so customers may not move if the gains from switching are small. Secondly, customers may be motivated by their opinions of the companies concerned. Electricity is a homogeneous product, but electricity retailers may need to be concerned about their corporate image. Thirdly, many customers may be unaware of the possibility of switching, or may have partial information regarding the potential benefits. Hence we might expect that marketing on the part of a regulator who makes this information available to customers may result in more fluid switching behaviour. This paper makes use of an extensive data set covering the North Canterbury region of New Zealand, provided by the local lines company, MainPower (see Section 2.2). The New Zealand electricity market separates retail and generation activities from ownership of distribution networks (the role of the lines company). Retailers compete for customers in a region, but the lines company has a monopoly on distribution. As such, MainPower observes all switching activity in the region. We use the entire population of ICPs

2 It should be noted that between the time of Brennan’s paper (2007) and this article (2015), the number of electricity switches in Britain has roughly halved, while in New Zealand, it has roughly tripled. Norway has also enjoyed quite high switching rates, growing from 8% in 2009 to 15% in 2014. Australian customers also exhibit high rates of turnover, with the state of Victoria experiencing 10.8% turnover in the June quarter of 2014. 3 Puller and West (2013) make similar observations for the US market, while Su (2015) notes that US price declines were concentrated in early years of deregulation rather than ongoing. The concept that obfuscation may be in the interests of market participants, has been shown in the marketing literature by Kalayci and Potters (2010).

327

(individual meters) in our study, avoiding the biases that affect empirical work where participants must “opt in” to the study. Existing empirical work on individual customer switching behaviour has been sparse. Giulietti et al. (2005) examine behaviour of gas customers in the United Kingdom, who face the opportunity of switching from British Gas to new entrants in the market. Giulietti et al. (2014) use a theoretical model of customer reluctance to switch to infer customer search costs from the dispersion of prices. Deller et al. (2014) examine the outcomes of “The Big Switch”, where participants were “auctioned” to retailers. Participants were then offered the opportunity to switch from their existing retailer to either one or two of the most competitive retailers. Closest to our work is that of Hortaçsu et al. (2015), who study the switching behaviour of Texas residents in the wake of deregulation. We examine the transfers between retailers of ICPs, as retail prices change. We also document switching behaviour following three events. First, we measure the response of customers to Contact Energy’s Directors’ remuneration. Contact Energy has a large retail presence in the MainPower region (see Fig. 3), as it is the incumbent retailer. In August 2008, Contact announced that it was raising retail prices.4 Following this, in September 2008, it announced a doubling of the remuneration of its Directors.5 The proposal was widely criticised in the media in October, during the lead up to Contact’s Annual General Meeting.6 Although the board was able to pass the motion, in the face of shareholder opposition, Contact decided to delay remuneration increases.7 Contemporary media coverage attributed subsequent competitor market share gains to disenfranchised Contact customers seeking a new retailer.8 We ask empirically whether this event was followed by increased switching activity among customers over and above price effects; were customers potentially concerned with the corporate governance of Contact? Second, we see a marketing campaign, in which Trustpower (a gentailer traditionally based in the North Island of New Zealand, which was expanding its presence in the South Island) marketed itself as being more desirable to customers on the strength of being a trust (as opposed to a company, like most other retailers). Do customer movements, following this campaign, indicate a potential preference over ownership structure for their retailer?9 Ethical concerns of customers are a controversial concept in the economics literature. On the one hand, Sen et al. (2001) argue that even in the presence of an organised boycott of a particular product, consumers have weak incentives to participate: at the margin, they are unlikely to achieve change, and potentially face high costs in terms of suboptimal consumption patterns to achieve this. On the other hand, Bénabou and Tirole (2006) argue that individuals may engage in ethical behaviour as a signalling tool. By engaging in costly activities, individuals signal that they may have preferences of a particular type, which allows them to relate to others with similar preferences. This is potentially an economically significant issue: Arnot et al. (2006) show that in the coffee market, fair trade coffee has a markedly lower price elasticity than other products, implying that under oligopolistic competition, ethical products may have higher profit margins.

4 “Contact Energy is warning that household power prices will rise at least 6% this year.” (Weir, 2008). 5 This was revealed in September 2008 in Contact’s announcement of its October Annual General Meeting. 6 “Prime Minister Helen Clark has hit out at Contact Energy’s ‘greed’ in planning to double directors’ fees while reaping money from the public through high power charges.” (Watt, 2008). 7 Janes (2008). 8 “TrustPower boosted customer numbers to 227,000 from 222,000 a year earlier helped by fallout from Contact Energy’s price rise and directors’ fees debacle and a push into Northland.” (Bradley, 2009). 9 The ownership structures of New Zealand electricity retailers are summarised in Appendix A.

328

T. Daglish / Energy Economics 56 (2016) 326–337

Third, we examine the introduction of the “What’s my number?” campaign by the Electricity Authority. In this campaign, the market regulator provided a website to help customers estimate the savings they could achieve by switching retailer. The desired effect of this intervention would be to lower search costs and thus achieve greater customer movement. Our study measures how large the change in turnover was, following the campaign. Lastly, we note that during the period of our study, a number of retailers entered the market. Work by Deller et al. (2014) shows that customers presented with multiple choices between competing electricity retailers are less likely to switch away from their current retailer. We test whether this holds in our sample, by allowing the number of retailers present in the market to affect consumer inertia. Surprisingly (but consistently with Deller et al. (2014)), we find that more retailers present in the market results in greater inertia, and less customer movement. The layout of the remainder of this paper is as follows. Section 2 outlines our methodology and data set. Section 3 presents the empirical results of our work, while Section 4 concludes. 2. Material and methods 2.1. Methodology Empirically, our work makes use of a Conditional Logit model for estimating responses to product features. Critical to this model are price and inertia factors. Since electricity offtake data (in this study) is only available at an aggregated level, we must also estimate a demand proxy for individual ICPs. 2.1.1. Conditional Logit model The main results in this paper make use of the Conditional Logit model (McFadden (1973)) to model ICP switching behaviour. We assume that ICP i in month t receives utility from retailer j according to the following equation Ui,j,t =



k bk X i,j,t,k

+ 4i,j,t ,

(1)

where Xi,j,t are a set of characteristics (indexed by k) that vary across ICPs, retailers and time. These are observable to the econometrician. The term 4i,j,t is a term that is unobservable to the econometrician, but is assumed to follow a Type I extreme value distribution. As such, we can evaluate the probability that an ICP chooses a particular retailer (j) as Pi,j,t = 



 k bk Xi,j,t,k  .  exp k bk Xi,j ,t,k j exp

(2)

With this set of probabilities, the parameters (bk ) can be estimated by maximising the likelihood that the population of ICPs choose the particular retailers observed in each month, i.e.: max



b1 ,...,bN

i

j

a

i,j,t Pi,j,t

t

where a i,j,t is one if ICP i chooses retailer j at time t, and zero otherwise. 2.1.2. Price and search cost effects We assume that ICPs have a preference for cheaper retail rates, so our first explanatory variable is relative price: Xi,j,t,1 =



pj,t a  i,j ,t−1 p j j ,t

,

i.e. the price of a competitor relative to the ICP’s current retailer. This effect is offset by search costs. Hence, we allow our second term to be a preference for staying with the ICP’s current retailer: Xi,j,t,2 =

ai,j,t−1 = 1

1

if

0

otherwise.

A high estimate for b2 therefore represents higher search costs, manifested as greater inertia among customers. By interacting demographic or time series variables with Xi,j,t,2 , we can examine the effect of variables that might increase/decrease inertia (such as the “What’s my number?” campaign). By interacting variables with Xi,j,t,1 we can examine variables that might increase a household’s sensitivity to prices (such as estimates of ICP offtake: see Section 2.1.3). Hortaçsu et al. (2015) highlight the importance of identification of inertia effects in a model, and point out that including firm dummies cannot explain the long periods that customers spend with individual retailers. In their model, they break the switching decision into a two-step process, where customers first make a decision to “shop around”, while in the second stage, customers choose a retailer from the set of competitors. In our case, we incorporate the first decision into our model by way of the current retailer dummy: most customers will have a substantial preference for their existing retailer in a given month, resulting in long periods with no switching for most customers. We believe that this achieves a similar outcome to the Hortaçsu et al. (2015) approach, and by interacting other variables with Xi,j,t,1 and Xi,j,t,2 , we can model customers becoming more or less active in the market. Where we feel our approach has a slight advantage is in interpretation: since the coefficients are being estimated in the same logit formulation, it is easy for us to interpret the economic significance of inertia, as demonstrated in Section 3.2. In Section 3.3, we check the robustness of our approach by considering a sub-sample of ICPs, restricted to be only those who switch at least once in the sample. 2.1.3. Proxies for electricity usage As noted in at the start of this section, while we observe individual ICP movements between retailers, we do not observe individual ICP electricity usage. Electricity usage is potentially an important input for switching choice, since offtake multiplied by price will determine bill size. Large bills can result in “bill shock”, a motivation for switching firms. To circumvent our data’s shortcoming, as a first step prior to estimating the model in Section 2.1.1, we estimate a model for electricity offtake using data at a higher level of aggregation. The preponderance of customers in the MainPower region can be described as either rural customers or suburban/urban customers. The suburban/urban customers are characterised by high electricity usage during winter for heating (air conditioning usage in this region in summer is minimal), while the rural customers will potentially irrigate during the summer months, leading to an opposite seasonal pattern. We define ICPs as rural or urban based on their meshblock categorisation (see Section 2.2.5). We can then use electricity offtake at the Grid Exit Point (GXP; see Section 2.2.4) level to estimate a time series for demand for rural or suburban/urban customers. Estimation proceeds as follows. Demand by ICP i at time t is assumed to be lognormally distributed: log Yit = ct(i) Xit −

1 2 s + 4it 2 t(i)

where t(i) denotes the “type” of ICP i (rural or suburban/urban), and 4it has standard deviation s t(i) . Since there are many ICPs connected to a given GXP (and denoting that set I ( j)), we then use the law of

T. Daglish / Energy Economics 56 (2016) 326–337

329

Auckland

Wellington CUL0661

CUL0331

WPR0331 SBK0331 WPR0661

ASY0111 KAI0001

Christ church

0

400 km 300 mi

0

0

0

20

10

40 km

20

30 mi

Legend ICP GXP cover ages NIWA climate station

Fig. 1. The MainPower region. Shading denotes topography (hills and mountains). Boxed and shaded portions show the regions whose ICPs are attached to a particular GXP. Dots are individual ICPs. Small circles show the location of weather stations, used for meteorological data for our study.

large numbers to define the approximate distribution of offtake from GXP j at time t: ⎛ Zjt ∼ N ⎝



exp(ct(i) Xit ),

i∈I ( j)



⎞   2 exp(2ct(i) Xit ) exp(st(i) ) − 1 ⎠

i∈I ( j)

2.2. Data

≡ N(l jt , g2jt ), where l jt and g2jt are the mean and variance of the GXP j offtake at time t, themselves a function of the data (Xit ) and parameters (crural , curban , s rural and s urban ). By observing Zjt across all GXPs and time periods, we can construct the log-likelihood function

L=

j,t

 log



We use this as a proxy for expected monthly electricity demand for a customer. Interacting this proxy with relative prices in the Conditional Logit model allows customers with larger monthly offtakes to be potentially more sensitive to price differentials than lower use customers.

2pg2jt



 2  1 Zjt − l jt . + 2 g2jt

(3)

We choose the model parameters (crural , curban , s rural , and s urban ) to maximise the log likelihood (3). Having estimated the parameters, we can then calculate the expected offtake by a given ICP as   E(Yit ) = exp ct(i) Xit .

Our data cover the period May 2007–December 2012. The data used in this project can be split into six subsets: switching, prices, satisfaction, electricity usage, demographics, and weather information. 2.2.1. Switching In New Zealand, the electricity market can be divided into four sectors: retail (selling to end customers), generation (producing electricity), transmission (moving electricity around the grid, between GXPs) and distribution (connecting customers to the grid). Of these, retail and generation are competitive markets, whereas transmission and distribution are monopolised. In the case of transmission, services are provided by Transpower (a State Owned Enterprise), whereas in the case of distribution, services are provided by a collection of “lines companies”, each of which is a local monopoly. Our

330

T. Daglish / Energy Economics 56 (2016) 326–337 26

4

3

x 10

25 2.5

24

Price (c/KWh)

Frequency

2

1.5

Contact Meridian Tiny Mighty Power Genesis TrustPower Mercury Pulse

23

22

21 1

20

19

0.5

0

18 2007 0

1

2

3 4 5 Number of switches

6

7

Fig. 2. Histogram of number of switches for an ICP in the MainPower region. Vertical axis shows number of ICPs that have switched the given number of times during our sample period.

switching information is provided by MainPower, the lines company who provides distribution services in North Canterbury. Data are anonymised by meshblock (the smallest unit of measurement of the New Zealand census, corresponding to roughly 50 households) and covers all connections and terminations. The advantage of dealing with data provided by a lines company is that the lines company is privy to all customer movements, rather than only those concerned with a particular retailer. Note that any issues of power outages or lines failures are the responsibility of the lines company (MainPower), not the retailer with whom the customer deals. As a result, reliability of service is homogeneous across retailers in a given region. Fig. 1 shows the dispersion of ICPs across the region, and provides an idea of the demographic character of North Canterbury. The region contains some urban areas in the South: Rangiora and Kaiapoi. There are also some smaller towns scattered across the region: Kaikoura, Waipara, and Culverden. The region is also characterised by large rural areas. North Canterbury has varied terrain; the region 0.9

0.85

0.16

Contact Meridian

0.8

TMP Genesis

0.75

Trustpower Marketing Campaign

0.14

Contact Director Incident

0.12

Trustpower

What’s 0.1 My Number?

Mercury Pulse

0.7

0.08

0.65

0.06

0.6

0.04

0.55

0.02

0.5

2008

2009

2010

2011

2008

2009

2010

2011

2012

Year

8

0

Fig. 3. Market shares (as portion of all ICPs in the market) for the MainPower region. Important dates for our study are marked with vertical lines. Contact market share is measured on the left axis, while other firm market shares are measured on the right axis.

Fig. 4. Prices offered by retailers for the MainPower region. Note that Kaiapoi customers, due to a slightly different contractual arrangement with the MainPower lines company, can, theoretically, be offered a different tariff to other customers. The prices listed here are for non-Kaiapoi customers; differences for Kaiapoi customers are minimal (Genesis Kaiapoi customers saved 2.15 c/KW h extra in May–August 2007, TrustPower Kaiapoi customers paid an extra 0.11 c/KW h from May–September 2010, and 0.12 c/KW h thereafter). Tiny Mighty Power was only available in Kaiapoi, so the prices listed here are for Kaiapoi customers.

is dominated by the Canterbury plain, but in the West is bounded by the Southern Alps, and in the North by the hills surrounding Kaikoura. Canterbury was affected by two major earthquakes during the period studied. The first earthquake, on 4 September 2010, measured 7.1 on the Richter scale (centred West of Christchurch). The second, on 22 February 2011, measured 6.3, and caused further damage, including 181 deaths (mostly in Christchurch city). These two earthquakes caused substantial migration in the Canterbury province (of which the MainPower region is a subset). While there was substantial emigration due to the earthquakes, demolition and reconstruction work led to immigration of tradespeople to work in and around Christchurch. The southern parts of the MainPower region (Kaiapoi and surrounds) were particularly affected by these flows. The region is divided into seven GXPs where each ICP is assigned to one GXP (see Fig. 1).10 Ashley11 includes a direct connection to the transmission network by a fibre board factory. We exclude Ashley11 from our work on electricity usage, since its offtake is dominated by the fibre board factory. There are 38,880 ICPs in the region. Working at a monthly frequency (and after data cleanup), we have 1,897,085 ICP-month observations in our data, while we observe 16,633 switches during the period. A histogram of number of switches observed for each ICP is given in Fig. 2. Our data set consists of ICPs, rather than specific retail customers. Hence one should be aware that what we term as a “switch” could be a customer moving house, and bringing his/her retailer with him/her. Two comments are in order: first, this extra “noise” biases us against finding patterns in the data. Therefore our findings in Section 3 are more robust. Second, from the perspective of retailers or gentailers participating in the market, it is actually the ICP affiliation that matters. A retail customer moving from one GXP to another changes the portfolio of electricity that the retailer must purchase in the wholesale market. Hence an analysis of ICP switching is in many ways more relevant than an analysis of customer switches per se.

10 Grid Exit Points form the “nodes” of the New Zealand Electricity Market’s nodal structure. Hence a retailer supplying two customers at the same GXP is purchasing electricity for each customer at the same price at any given point in time.

T. Daglish / Energy Economics 56 (2016) 326–337

Culverden33

331

Culverden66/Kaikoura33

7000

3500

6000 3000 5000 4000

2500

3000 2000 2000 1000 2007

2008

2009

2010

2011

2012

1500 2007

2008

Waipara33

2009

2010

2011

2012

2011

2012

2011

2012

Waipara66

2000

4000

1800 3000 1600 1400

2000

1200 1000 1000 800 2007

2008

2009

2010

2011

2012

0 2007

2008

Kaiapoi11

2009

2010

Southbrook33

3500

3000 2800

3000 2600 2500

2400 2200

2000 2000 1500 2007

2008

2009

2010

2011

2012

1800 2007

2008

2009

2010

Fig. 5. Electricity offtakes by Grid Exit Point. Prior to November 2007, Waipara33 did not exist, and its ICPs were connected to Waipara66. Given this change in configuration, we exclude the pre-November 2007 data for Waipara33 and Waipara66 in our electricity demand estimation.

Contact Energy is the incumbent firm from deregulation, and, at the start of the sample, accounts for roughly 90% of ICPs in the region. Over the course of the sample period, this portion declines and the retail shares become more evenly distributed across retailers (see Fig. 3). In particular, note that Contact market shares decline significantly following the Trustpower Marketing campaign, and there is a marked convergence (small participants gain market share, while Contact declines) of market shares following the “What’s my number?” campaign. 2.2.2. Electricity prices Electricity retailers are required to report their “low usage” tariff to the Ministry of Economic Development.11 The low usage tariff is a plan that all retailers must offer, and generally consists of a relatively high kilowatt-hour price, combined with a relatively low monthly fee. In reality, most retailers offer a range of plans, differing on such factors as: discounts for prompt payment, combined gas and electricity provision, or even smoothing of costs over the year. Further, customers may, as a result of promotions, be given preferential rates as new customers, or as an inducement to not switch power

11 The Ministry of Economic Development was amalgamated into the Ministry of Business, Innovation and Employment on 1 July 2012.

companies. This information is not available (being bilateral arrangements between the retailer and the customer). The low tariff series is easily comparable across all retailers, so we use this as a proxy for the prices that customers faced when comparing retailers.12 Fig. 4 shows the time series of the low cost plan price for each retailer. As can be seen, the prices have increased almost monotonically over the period. Most retailers update their prices once or twice a year. Although the prices are correlated, there is clearly some dispersion, with some firms being slow to follow the trend, particularly toward the end of the data period. 2.2.3. Satisfaction Quality of service can be an important factor in customer satisfaction. While quality of delivery of electricity in New Zealand should be independent of retailer choice (since distribution and transmission are provided by the local lines company and Transpower, respectively), the Electricity and Gas Complaints Commissioner reports that for the 2010–2011 year, 40.3% of complaints regarding retailers concerned billing, and 19.4% concerned customer service. Billing

12 In general, our measure of price understates the per-unit charge paid by customers. However, since we are concerned with relative prices, if the biases are of similar size, the effects should cancel out. We are indebted to Stephen Burrough from PowerSwitch, who helped us identify specific dates when prices changed.

332

T. Daglish / Energy Economics 56 (2016) 326–337

Table 1 Summary statistics for demographic and weather data. Part A shows proportions aggregated to GXP level. GXP proportions are found by weighting meshblock proportions by the number of ICPs in that meshblock who are connected to the particular GXP. Mean, minimum, maximum and standard deviation are across all ICP-month data, inferred from the meshblock that the ICP resides in (although demographic information does not vary over time, not all ICPs are present at all points in time). Density is calculated by dividing the number of ICPs (in thousands) by the meshblock surface area in km2 . Part B illustrates weather statistics, which are calculated across all meshblock-months used in the data. Rain is measured in millimetres. Soil moisture is moisture as percentage of soil capacity. Temperature is in degrees Celsius. All weather variables except sunshine hours are averaged across the month from daily observations. Sunshine is measured in hours across the month. A. Demographic data Location

ICP density

Prop 4 bedrooms

Prop 100K + income

Prop age 65+

Culverden33 Culverden66 /Kaikoura33 Waipara33 Waipara66 Kaiapoi11 Southbrook33

0.29

24.06

14.10

10.67

0.64 0.31 0.48 0.19 0.09

23.99 21.66 29.85 27.78 28.59

14.21 10.42 17.02 11.29 11.01

14.65 15.60 13.67 14.98 15.54

Mean Min Max Std. dev.

0.43 0.00 2.04 0.49

27.48 0 77.78 13.66

14.80 0 60.00 11.10

14.07 0 64.29 9.34

Rain

Sunshine

Soil moisture

Temperature

63.61 2.37 463.18 44.89

178.14 80.00 295.38 46.50

28.59 9.51 58.36 10.65

11.50 2.61 19.57 3.74

B.Weather data

Mean Min Max Std. dev.

and customer service practices could vary considerably across retailers. To proxy for this effect, we use the ConsumerNZ annual Energy Provider’s survey scores. This national survey targets ConsumerNZ subscribers, and asks their opinion of their current retailer. The response rate for this survey over the 2008–2011 period ranges from 8,448 to 13,047. Since this series is annual, we assign the values from the survey to all months occurring in that year. Dispersion of these scores is considerable, with a mean firm-year satisfaction of 69.5% and standard deviation 7.2%.

urban customers to perform our estimation of rural/urban demand as outlined in Section 2.1.3. Summary statistics are provided in Table 1. Demographic variables are reported as averages for the different GXPs, and as a distribution over our ICP-month data. There is no time-series component to the census information (censi are held every six years). However, for the ICP-month data, the number of ICPs in a particular meshblock may fluctuate due to missing observations or addition of new GXPs, although these effects are minimal.14

2.2.4. Electricity usage Electricity offtakes are available at a GXP level (see Fig. 1). We aggregate daily offtakes into monthly observations. Electricity use fluctuates in most areas and often shows strong seasonal patterns. These patterns can exhibit high mid-year (winter) usage (see, for example, Kaiapoi11) or high year end (summer) usage (see, for example, Waipara33). Graphs of the time-series behaviour of electricity offtakes are contained in Fig. 5.

2.2.6. Weather Weather information is available from the National Institute of Water and Atmospheric Research (NIWA). NIWA has a range of weather stations scattered across the North Canterbury region (see Fig. 1). Each weather station reports a range of weather characteristics, such as temperature, hours of sunshine, rainfall, and soil moisture. For each ICP location, we take the weather reading from the closest weather station. We then generate an average across all ICPs in a particular meshblock to generate meshblock level weather parameters. These meshblock level weather numbers are used for all ICPs within the meshblock.15 We summarise this information in Table 1.

2.2.5. Demographics Since our data are anonymised at a meshblock level, we can link our switching data to Statistics New Zealand data concerning house sizes, and general demographic information for the area. These data consist of portions of residents in a meshblock who conform with a particular characteristic. For example, we see the portion of people who live in a house with four bedrooms (almost all houses in the region have 3 or 4 bedrooms), or the portion of people who are age 65+. We also observe the surface area of a given meshblock, and this, combined with our ICP positions, allows us to construct an ICP density measure, which we use as a continuous measure of rural/urban mix. In our demand estimation, we assume that all residents of a given meshblock are rural or urban according to Statistics New Zealand’s definitions.13 We then count the total GXP population of rural and

13 We categorise “Major Urban Area”, “Satellite Urban Area”, “Independent Urban Area” or “Rural Area with High/Moderate Urban Influence” as urban. “Rural Area with Low Urban Influence” or “Highly Rural/Remote Area” are classified as rural.

3. Results We first investigate the electricity offtakes in the MainPower/North Canterbury region. Then we turn our attention to the actual switching decisions of the ICPs. 3.1. Electricity demand We use our panel data on GXP offtakes, weather, and rural/urban information to explore the sensitivity of rural/urban electricity

14 In our calculation of GXP level demographics, we work with the full set of ICPs, effectively ignoring time series characteristics, for this reason. 15 Since our ICP level switching data is anonymised, we cannot link ICP level weather conditions with our switching data, and thus we must aggregate weather data to a meshblock level.

T. Daglish / Energy Economics 56 (2016) 326–337 Table 2 Dependent variable is log offtake per ICP in a given month. Standard deviation gives the standard deviation of the customer type’s log monthly offtake. Parameters are estimated as described in Section 2.1.3. Rural Variable Rain Sun Soil moisture Temperature Dummy Jan Dummy Feb Dummy Mar Dummy Apr Dummy May Dummy Jun Dummy Jul Dummy Aug Dummy Sep Dummy Oct Dummy Nov Dummy Dec Standard deviation

Urban

Coeff.

T-stat

Coeff.

T-stat

0.0001 0.0006 −0.0081 −0.0038 8.3322 8.1027 7.9772 7.7255 7.5759 7.6357 7.7599 7.7378 7.6458 7.7918 8.1537 8.2805

0.1036 0.2671 −0.7381 −0.2205 13.8501 15.0269 14.3355 16.0595 14.8771 13.8339 14.1722 12.9520 11.7069 11.6837 13.3471 15.0311

0.0002 0.0002 0.0010 0.0039 7.4483 7.3810 7.4912 7.5465 7.7043 7.8073 7.8676 7.8075 7.6339 7.5749 7.4969 7.4798

1.3257 0.9685 0.7607 2.0193 97.2585 110.9690 110.5993 122.4280 110.4374 108.8133 96.7127 90.3768 88.1384 87.3372 93.5722 94.3005

0.4855

16.3583

0.0329

8.2382

demand to time of year and weather. As described in Section 2.1.3, we estimate this model by dividing customers (by meshblock) into rural and urban. We then choose demand parameters so as to maximise the likelihood of these demand functions generating the GXP level demand observed. Table 2 contains the results of the estimation of rural and urban demand as a function of rainfall, sunshine hours, soil moisture, temperature, and month of the year. In both cases, a strong seasonality is present. For urban ICPs, electricity usage peaks during the winter months (June–August). For rural ICPs, the drier summer months represent the high-use period. In both cases, weather effects are small. We note (statistically insignificant) negative effects from soil moisture and temperature on rural offtakes. Urban offtakes are significantly positively affected by temperature, although this effect is dominated by the seasonal patterns. An ICP’s demand (for use in the Conditional Logit model) is estimated using its meshblock weather variables to evaluate the rural or urban model, depending on whether it lies in a rural or urban meshblock. 3.2. Switching behaviour We next turn our attention to customer switching behaviour. In our estimation of standard errors for the Conditional Logit model, we allow for clustering across ICPs. As a robustness check, we also calculate standard errors by clustering over meshblocks.16 Our approach to clustering follows Cameron et al. (2011). To test whether there is “bill shock”, we include our proxy for electricity demand (see Sections 2.1.3 and 3.1). We interact the demand proxy with relative pricing, so that customers could respond to a large bill by switching to a cheaper retailer. Given Contact’s position as regional incumbent, we also include dummy variables for each of the other retailers, effectively allowing customers to exhibit a preference for a particular firm over others. Table 4 contains our results. Our basic finding (before examining the events) is that customers have a very strong preference for

16 Clustering over ICPs is potentially important, since a household occupying an ICP may have a (unobservable) preference for particular firms due to past experiences. Clustering over meshblocks may prove important since our demographic information is only available at the meshblock level (see Section 2.2.5). This results in less intra-meshblock variation than might be the case if we saw individual household level demographic information.

333

Table 3 Description of explanatory variables for switching model. Variable

Description

CurrentR RP Director N

Dummy variable: 1 for current retailer. Retailer price, divided by current retailer price. Dummy variable: 1 for Contact N months following Director incident (August 2008). Director Dummy variable: 1 for Contact from September 2008–August 2009. Return N Dummy variable: 1 for Contact N months following Director incident, if customer was with Contact at time of Director incident, and is currently with another retailer. Return Dummy variable: 1 for Contact from September 2008–August 2009, if customer was with Contact at time of Director incident, and is currently with another retailer. TPTP N Dummy variable: 1 for Trustpower N months after Trustpower campaign (February 2010). TPTP Dummy variable: 1 for Trustpower from March 2010–February 2011. TPCT N Dummy variable: 1 for Contact N months after Trustpower Marketing campaign. TPCT Dummy variable: 1 for Contact from March 2010–February 2011. WMN? Dummy variable: 1 for current retailer after “What’s my number?” campaign (June 2011). NRetailers Number of retailers present in market, interacted with CurrentR. 9/2010EQ Dummy variable: 1 for current retailer after first earthquake (September 2010). 2/2011EQ Dummy variable: 1 for current retailer after second earthquake (February 2011). Satisfaction Satisfaction percentage, as decimal (e.g. 50% = 0.5), interacted with CurrentR. Den Thousands of ICPs per km2. 4Bed Proportion of houses within ICP’s meshblock that have 4 bedrooms, as decimal. Inc Proportion of households within ICP’s meshblock that have $100,000 or more income per annum, as decimal. Age Proportion of population within ICP’s meshblock that are age 65+, as decimal. Off Estimated offtake for this ICP/month (see Section 2.1.3), in GWh. Meridian Dummy variable: 1 for Meridian. TMP Dummy variable: 1 for Tiny Mighty Power. Genesis Dummy variable: 1 for Genesis. Trustpower Dummy variable: 1 for Trustpower. Mercury Dummy variable: 1 for Mercury. Pulse Dummy variable: 1 for Pulse.

remaining with their current retailer. With a coefficient for current retailer of 5.9894, we can infer that with seven firms, if all prices were identical, and there were no firm specific preferences, a cus1 tomer has probability e5.9894 = 0.0025 of switching to a given +6 competing retailer each month. Our coefficient of −17.5015 on price says that if a customer is offered a 10% cheaper price to switch 0.1×17.5015

retailer, he/she would have probability e5.9894e+e0.1×17.5015 +5 = 0.0140 of switching; price competition has significant impact on customer churn. These numbers, when compared to other papers studying the UK and US, support the aggregate level observation that New Zealand has high switching rates. 17

17 Giulietti et al. (2014) estimate a cash value of search costs in the UK. For a 3300 KWh per year user, they estimate (for 2005) a lower quartile for search costs of £23.13. Our incumbent effect (as a fraction of prices) is 34.22%. A MainPower consumer using 3300 KWh per year, and paying 22c per KWh would have an implied (per month) 5.9894 3300×0.22 ). Hortaçsu et al. (2015) estimate a search cost of $NZ 20.70 (= 17 .5015 × 12 propensity to search for a new supplier, and then a Conditional Logit model for selecting a new supplier. Repeating our thought experiment of a single firm (among seven) lowering prices by one cent (roughly 10%), their estimates imply that, at the start of their sample period (January 2002), a new firm would have a 0.12% chance of stealing a customer from the incumbent retailer, and a 0.23% chance of stealing from another entrant. If the firm lowering prices was the incumbent, then it would have a 2.65% chance of stealing a customer from an entrant. In contrast, two years into their sample, the incumbent effect declines, so that an entrant would have a 0.31% chance of taking a customer from the incumbent, and a 0.56% chance of taking a customer from another entrant. During this later period, if the incumbent reduced its prices, it would have a 1.32% chance of taking a given customer from an entrant.

334

T. Daglish / Energy Economics 56 (2016) 326–337

Table 4 Results for customer switching behaviour. T-stats are constructed using standard errors that are robust to errors clustered by ICP (t(I)) and meshblock (t(M)). Variables are described in Table 3. Variable

Coeff.

t(I)

t(M)

5.9894 −17.5015

199.0698 −111.6189

145.2734 −92.0877

Director Remuneration Director 1 Director 2 Director 3 Director 4 Director 5 Director 6 Director 7 Director 8 Director 9 Director 10 Director 11 Director 12 Dir * Den Dir * 4Bed Dir * Inc Dir * Age

0.1449 0.3090 0.2689 −0.8798 −1.1990 −0.4826 −0.8228 0.1700 −0.0334 0.0016 0.2299 −0.1984 −0.1890 0.0048 −0.0054 0.0113

0.8235 1.6785 1.4852 −6.4479 −9.6355 −3.4164 −6.1735 1.1002 −0.2171 0.0104 1.4087 −1.3980 −2.6123 1.5830 −1.4568 3.1659

0.6415 1.3397 1.2767 −4.5047 −6.6982 −2.3096 −3.8767 0.6724 −0.1770 0.0083 1.1486 −0.9488 −2.0622 0.9333 −0.9527 1.9379

Contact recovery of lost customers Return 1 Return 2 Return 3 Return 4 Return 5 Return 6 Return 7 Return 8 Return 9 Return 10 Return 11 Return 12 Return * Den Return * 4Bed Return * Inc Return * Age

– 0.4584 −0.2938 0.4301 0.0382 −2.0275 −1.0543 −1.0063 −0.4434 −1.0427 −0.7180 −0.5420 0.9766 0.0106 −0.0076 −0.0066

– 0.3375 −0.2573 0.5613 0.0530 −1.7572 −1.3226 −1.5817 −0.7816 −1.7781 −1.4265 −0.9237 3.1840 0.6747 −0.3077 −0.3789

– 0.3106 −0.2497 0.5371 0.0504 −1.6939 −1.2814 −1.5738 −0.7351 −1.6844 −1.4094 −0.8613 3.1408 0.6418 −0.2839 −0.3171

Inertia effects WMN? WMN? * Den WMN? * 4Bed WMN? * Inc WMN? * Age NRetailers 9/2010EQ 2/2011EQ Satisfaction

0.1351 −0.4582 −0.0019 −0.0009 0.0051 0.2135 −0.1650 −0.2124 1.2353

1.6642 −8.9615 −0.8887 −0.3400 2.0032 20.8788 −4.8149 −7.1507 40.9681

1.2805 −6.8379 −0.6375 −0.2536 1.5551 19.2636 −4.1002 −6.0443 21.7922

Firm effects Meridian TMP Trustpower

−0.6926 −0.6641 −2.2723

−34.0550 −8.5519 −61.9556

−22.7614 −5.6708 −37.1430

CurrentR RP

When examining the events in our time period, we check for a delayed effect on customer behaviour by including time dummies for the twelve subsequent months following an event. We further interact this twelve-month period with demographics to see which groups switched during the time period considered. Examining the Director Remuneration incident, we find fairly weak effects in terms of customer movements. During the early months (September–November 2008), Contact in fact seemed to have better than usual retention/attraction of customers. However, December 2008 to March 2009 saw an outflow of customers. Indeed in January, with coefficient −1.1990, we conclude that the Director Remuneration effect was the equivalent of a 6.9% price differential for isolated regions (as ICP density increases, the

Variable

Coeff.

t(I)

t(M)

Trustpower campaign effect: Trustpower TPTP 1 TPTP 2 TPTP 3 TPTP 4 TPTP 5 TPTP 6 TPTP 7 TPTP 8 TPTP 9 TPTP 10 TPTP 11 TPTP 12 TPTP * Den TPTP * 4Bed TPTP * Inc TPTP * Age

−1.8433 0.0607 1.4865 0.8561 −0.7251 −0.6937 −0.9064 −0.4990 −0.5657 −0.6925 −0.9875 −0.6673 0.8778 0.0109 0.0072 0.0043

−7.7302 0.4393 11.4606 6.1175 −3.9584 −3.6976 −5.1949 −3.4922 −2.3133 −2.7412 −5.4123 −2.8577 11.8440 3.4544 1.8304 1.1801

−5.5148 0.2110 5.6245 2.7910 −2.2165 −2.2964 −3.2426 −1.7920 −1.6019 −1.8714 −3.3457 −1.9965 5.8672 1.5539 0.8573 0.5120

Trustpower campaign effect: Contact TPCT 1 TPCT 2 TPCT 3 TPCT 4 TPCT 5 TPCT 6 TPCT 7 TPCT 8 TPCT 9 TPCT 10 TPCT 11 TPCT 12 TPCT * Den TPCT * 4Bed TPCT * Inc TPCT * Age

0.6003 −0.9315 −1.3408 −0.7538 0.2909 0.4509 0.1467 0.3835 0.8371 0.6987 0.2455 1.6877 0.1411 0.0015 −0.0031 0.0055

4.8132 −8.6966 −13.1932 −6.8548 2.5783 3.7898 1.3699 3.7327 5.8479 4.6246 2.3031 11.9288 2.4691 0.6145 −1.0990 1.9693

2.9064 −4.6914 −7.4284 −3.2324 1.2814 2.2977 0.8379 2.0750 4.2209 3.3114 1.3221 8.4452 1.6188 0.2731 −0.5808 1.3064

Price effects Den * RP 4Bed * RP Inc * RP Age * RP Off * RP Inc * Off * RP

0.0262 0.1489 −0.5625 0.3648 −0.0249 0.1297

1.7336 2.3096 −1.6209 5.2313 −1.1520 0.8995

1.3070 2.0303 −1.5650 4.6373 −1.1526 0.8748

Genesis Mercury Pulse

−1.1708 −0.5533 −0.9098

−39.2721 −10.9371 −9.4215

−25.8757 −7.8094 −5.2020

ICP density interaction term would mitigate this effect slightly). Contact’s approaching of former customers to “win them back” was only effective in months 2, 4, and 5 for very rural areas. However, for higher ICP densities, more customers were seen to return. Given that these months follow the largest exodus month (January 2009) this may be seen to be the period in which the Director Remuneration effect is strongest. In addition to these effects, the Contact Satisfaction score fell by 7% from 2008 to 2009. Next, we turn our attention to the various marketing campaigns held in the region. The “What’s my number?” campaign has a small positive coefficient (indicating an increase in incumbent effects), suggesting that it has little effect on customer loyalty for sparsely

T. Daglish / Energy Economics 56 (2016) 326–337

335

Table 5 Results for customer switching behaviour, where sample is restricted to customers who switched at least once during the sample period. T-stats are constructed using standard errors that are robust to errors clustered by ICP (t(I)) and meshblock (t(M)). Variables are described in Table 3. Variable

Coeff.

t(I)

t(M)

6.2101 −18.1507

188.3477 −117.5283

135.8469 −100.3177

Director Remuneration Director 1 Director 2 Director 3 Director 4 Director 5 Director 6 Director 7 Director 8 Director 9 Director 10 Director 11 Director 12 Dir * Den Dir * 4Bed Dir * Inc Dir * Age

−0.2608 −0.0906 −0.1323 −1.2853 −1.6714 −0.9661 −1.3152 −0.1089 −0.3186 −0.2871 −0.0609 −0.4953 0.3563 0.0084 0.0000 0.0073

−1.4747 −0.4900 −0.7268 −9.3432 −13.2935 −6.7794 −9.7742 −0.7034 −2.0606 −1.8935 −0.3721 −3.4771 4.8995 2.7340 0.0031 2.0473

−1.1660 −0.4002 −0.6316 −6.5542 −9.2855 −4.7388 −6.3091 −0.4514 −1.6802 −1.5201 −0.3065 −2.4020 3.9395 1.7635 0.0021 1.1797

Contact recovery of lost customers Return 1 Return 2 Return 3 Return 4 Return 5 Return 6 Return 7 Return 8 Return 9 Return 10 Return 11 Return 12 Return * Den Return * 4Bed Return * Inc Return * Age

– 1.9361 1.2566 1.9765 1.5139 −0.5469 0.4346 0.2479 0.8075 0.2080 0.5315 0.7069 0.9556 0.0106 −0.0075 −0.0064

– 1.3875 1.1001 2.5802 2.1020 −0.4730 0.5464 0.3908 1.4274 0.3553 1.0588 1.2082 3.1246 0.6723 −0.3057 −0.3649

– 1.2776 1.0673 2.4655 1.9906 −0.4554 0.5266 0.3890 1.3384 0.3363 1.0376 1.1187 3.0662 0.6392 −0.2837 −0.3043

Inertia effects WMN? WMN? * Den WMN? * 4Bed WMN? * Inc WMN? * Age NRetailers 9/2010EQ 2/2011EQ Satisfaction

−0.2844 −0.1293 −0.0011 −0.0005 −0.0004 0.1936 −0.3886 −0.4122 0.2447

−3.4616 −2.5116 −0.4833 −0.1725 −0.1460 18.3639 −10.7987 −13.5656 6.9128

−3.0250 −2.1022 −0.3836 −0.1355 −0.1271 17.0370 −9.2259 −11.0327 5.1497

Firm effects Meridian TMP Trustpower

0.0425 −0.0740 −1.6954

1.9474 −0.9113 −43.8147

1.2848 −0.6415 −24.6463

CurrentR RP

populated areas.18 However, we see a negative coefficient for ICP density interacted with “What’s my number?”, suggesting a slight reduction in inertia for urban customers. Interestingly, the largest effect is due to the Trustpower campaign, starting in February 2010, where Trustpower markets itself as being New Zealand owned and being more desirable due to its trust governance structure.19 At its height (month 3) this campaign gives Trustpower the equivalent of an 8.5% price differential. Conversely,

18 This campaign’s effect appears to have been mostly on more actively switching customers (see Section 3.3). 19 See Appendix A for details of retailer characteristics.

Variable

Coeff.

t(I)

t(M)

Trustpower campaign effect: Trustpower TPTP 1 TPTP 2 TPTP 3 TPTP 4 TPTP 5 TPTP 6 TPTP 7 TPTP 8 TPTP 9 TPTP 10 TPTP 11 TPTP 12 TPTP * Den TPTP * 4Bed TPTP * Inc TPTP * Age

−1.7954 0.0017 1.5079 0.8899 −0.7080 −0.6793 −0.8964 −0.4633 −0.5517 −0.6912 −0.9310 −0.6422 0.8852 0.0098 0.0072 0.0053

−7.5466 0.0126 11.7131 6.3948 −3.8766 −3.6320 −5.1632 −3.2592 −2.2665 −2.7502 −5.1281 −2.7687 12.1093 3.1327 1.8485 1.4661

−5.5602 0.0063 5.9399 3.0224 −2.2407 −2.3251 −3.3472 −1.7368 −1.6122 −1.9294 −3.2816 −1.9701 6.1383 1.4282 0.8825 0.6658

Trustpower campaign effect: Contact TPCT 1 TPCT 2 TPCT 3 TPCT 4 TPCT 5 TPCT 6 TPCT 7 TPCT 8 TPCT 9 TPCT 10 TPCT 11 TPCT 12 TPCT * Den TPCT * 4Bed TPCT * Inc TPCT * Age

0.5751 −1.2061 −1.7358 −1.1696 0.0645 0.2847 −0.1176 0.1660 0.9242 0.7393 0.1456 1.7667 0.6690 0.0043 0.0007 0.0019

4.6620 −11.3934 −17.2987 −10.7636 0.5785 2.4177 −1.1121 1.6463 6.4738 4.9054 1.3865 12.5571 12.0551 1.8433 0.2501 0.6956

2.8956 −6.4078 −10.1191 −5.2660 0.2954 1.5145 −0.6840 0.9354 4.7537 3.6056 0.8145 9.0793 7.8276 0.8824 0.1388 0.4371

Price effects Den * RP 4Bed * RP Inc * RP Age * RP Off * RP Inc * Off * RP

−0.0480 −0.2435 −0.4196 0.1558 −0.0528 0.2511

−3.2345 −4.0137 −1.2125 2.4165 −2.4986 1.7475

−2.9243 −3.7578 −1.2954 2.0724 −2.8410 1.9326

Genesis Mercury Pulse

−0.6158 0.0692 −0.1874

−19.6607 1.2648 −1.8443

−12.3813 0.9351 −1.0165

the campaign is characterised by negative effects on Contact. Graphically, the results of this campaign can be seen in Fig. 3: from February 2010 onward, each winter, Trustpower market shares rise, largely at the expense of Contact market shares. Examining demographics, urban customers were on the one hand the most likely to move to Trustpower in the campaign, but were also the least likely to leave Contact (the interaction terms with ICP density for TPTP and TPCT are both positive). The number of retailers has a negative effect on switching in the market (consistent with the findings of Deller et al. (2014), where more choice deters customer switching). Quantifying this effect is slightly more challenging due to the increase in retailers changing the denominator of Eq.(2). However, we can infer that growing from

336

T. Daglish / Energy Economics 56 (2016) 326–337

five retailers to six retailers would allow existing retailers to raise their prices by 0.2% without losing market share, while growing from six to seven would admit price rises of 0.3%.20 The two earthquakes have the expected effect on switching behaviour. Noting that considerable migration was caused by the two earthquakes, we might expect to see more switching, caused by customers moving into and out of the region. Both earthquakes cause a significant decrease in inertia effects, suggesting that this was indeed the case. Satisfaction plays a modest role in determining loyalty to a customer’s current retailer. A 10% change in satisfaction, can be equivalent to a 0.7% price differential. Our company control variable coefficients are not surprising. Each of the non-incumbent firms has a negative coefficient. Interestingly, Trustpower’s number is the most negative, reflecting its small market share prior to the marketing campaign. The effect of the campaign is to largely offset this, resulting in Trustpower becoming the most popular firm during month three of the campaign. Examining demographic effects, we find that urban households, large households, and elderly households exhibit less sensitivity to prices, while high income households seem more sensitive. These findings echo the survey evidence of Giulietti et al. (2005). We find little evidence of bill shock on the whole, with our electricity offtake series having little explanatory power. Interacting this with income suggests that while there may be a small bill shock effect for low income households, higher income households generally move retailers in response to price differentials. 3.3. Movers One concern with our work would be that many ICPs do not switch during the sample period. This could raise an identification issue, in that with little movement, it would be difficult to distinguish between preference for the retailer who happens to be an ICP’s retailer at the start of the sample, and preference for the ICP’s current retailer, since these would be one and the same for non-switching ICPs. To examine whether this affects the results, we restrict our sample to ICPs who have switched at least once in the sample period, effectively excluding “non-movers”. In addition to addressing this possible identification issue, the movers are potentially interesting in their own right. As discussed in Hortaçsu et al. (2015), inertia can stem from customers either choosing not to investigate alternative retailers, or by investigating, and then choosing to remain with their current retailer. By focusing our attention on those ICPs where at least one switch has occurred, we create a sample that is more likely to have investigated switching over time. We might thus expect to find our estimates of sensitivity to external events to be stronger than for the pooled group, in which there could be some customers who are unresponsive (due to never investigating their options). Restricting our attention to “movers”

20 Market share neutral price growth, as the market grows from N0 to N1 retailers, is the solution of

ebCurrentR +N1 bNRetailers ebCurrentR +N0 bNRetailers = b ebCurrentR +N0 bNRetailers + N0 − 1 e CurrentR +N1 bNRetailers + (N1 − 1)e−bRP Dp where bCurrentR is the current firm coefficient, bNRetailers is the number of retailers interacted with current firm coefficient, and bRelPrice is the price sensitivity coefficient. Solving for Dp gives: Dp =

1 [−bNRetailers (N1 − N0 ) + log(N1 ) − log(N0 )] . bRP

Growing from 4 to 5 retailers would force existing retailers to lower prices to retain market share, since log(5) − log(4) > bN Retailers = 0.2135, and bRP = −17.5015 < 0.

reduces our sample size to 14,687 ICPs, with 795,933 ICP-month observations. The results are reported in Table 5. Our basic finding regarding the strength of the inertia effect is robust. Our coefficients for current retailer and relative price change from 5.9894 to 6.2101 and −17.5015 to −18.1507, respectively. While both coefficients’ differences are statistically significant, due to our large sample, the coefficients are numerically similar. Repeating our calculation of the effect of a firm giving a 10% price cut, we find that the probability 0.1×18.1507 of attracting an ICP is e5.9894e+e0.1×18.1507 +5 = 0.0150. Not surprisingly, conditioning on ICPs that change hands, this sensitivity is slightly higher, however it remains qualitatively similar. Our examination of the Director Remuneration period becomes more clear-cut. All months have negative coefficients, while all bar month 6 (February 2009) have positive coefficients for returning. Urban customers (high density) were less likely to leave, and more likely to return. In contrast, in the full sample, urban customers were more likely to leave Contact during this period. The “What’s my number?” campaign has a much stronger effect on this group than on the population as a whole, suggesting that the campaign encouraged increased mobility among already mobile customers. Again, this effect is stronger with urban customers. Trustpower’s marketing campaign gives qualitatively similar results to the pooled sample. Both earthquakes had more significant effects on this group.21 Satisfaction has a considerably lower effect on the Mover subgroup than on the population as a whole, suggesting that this group is less concerned about quality of service, relative to customers in general. Interestingly, we see some changes in the demographic effects for this sub-sample. Urban users are more price sensitive (in contrast to rural users in the overall sample). Larger households are also more price sensitive in this sample. Age and income effects, however, remain the same: high income households are more price sensitive, and older households are less sensitive. Both age and income effects are less pronounced in the mover subsample. This sheds some light on the population of movers versus non-movers. Movers with large houses, and who live in urban areas are found to be more price sensitive. In contrast, in the whole population, people with small houses, and who live in rural areas are more price sensitive. This suggests that non-movers may often be people with large houses who live in urban areas (whose inactivity offsets the behaviour of their moving subset). Regardless of the sample, older individuals are less price sensitive, suggesting that these contribute both to the non-mover population, and also are less price sensitive even when mobile. Our dummy variables show that Contact has a less privileged position relative to other firms with this group of customers. Meridian, Tiny Mighty Power, Mercury, and Pulse all have coefficients that are insignificantly different from zero (i.e. they are not statistically significantly different to Contact). 4. Conclusion This paper examines switching sensitivity to governance aspects of retailers/gentailers, over and above responsiveness to price differences. We find that there is a small response by customers to news about Contact’s Directors’ compensation. However, more significant customer movements occur in response to Trustpower’s marketing of itself as having a more desirable governance structure than its competitors. This campaign may well have

21 A customer moving due to the earthquake and precipitating a switch would automatically put the ICP in the “movers” group. Hence the greater effect of the earthquakes on this subsample is not surprising.

T. Daglish / Energy Economics 56 (2016) 326–337

been more effective because it was preceded by the remuneration incident. Our results suggest that inertia is a powerful force in the market, and that government sponsored initiatives such as “What’s my number?” can help to break this effect, albeit being more effective for those customers who have already decided to be mobile. If a new campaign were to be run, attempting to target rural areas could result in further uptake (the existing campaign was empirically least effective here). In addition, our results suggest that elderly customers are less price sensitive than younger customers, implying that retirees could be another group who could be targeted to improve market turnover. Interestingly, unlike previous studies, we find limited support for bill shock in motivating households to move. Households may be moving in response to lower price offerings even during low bill seasons. When we focus our attention on the subgroup of ICPs where at least one switch has taken place in our sample, we find stronger evidence of responses to Contact’s Director’s compensation, but weaker

337

satisfaction effects. Our tentative conclusion is that the active switchers are more concerned with price or taking a moral stance than responding to poor service. In some ways, our findings are reassuring from a policy perspective: consumers are searching for low prices, and are doing this consistently, rather than in response to high bills during the winter. On the other hand, the large portion of non-movers (see Fig. 2), and the weakness of the “What’s my number?” effect for the overall sample, raise some concerns about whether all participants are shopping around. Further, the increased inertia as more participants enter themarket suggests that there is confusion among some consumers. We believe that our methodology has considerable scope for application to other regions or other countries where customers face retail choice for electricity. Further work examining the market-wide effects of customer switching will help to understand the dynamics of this vertically integrated industry.

Appendix A. Summary of retailer characteristics

Table A1 Summary of retailer vertical integration and ownership structures. Retailer

Vertical integration

Ownership structure

Contact Energy Meridian Energy

Gentailer Gentailer

Tiny Mighty Power Genesis Energy

Retailer (owned by Gentailer) Gentailer

Trustpower Mercury Energy

Gentailer Gentailer

Pulse Energy

Retailer

Limited Liability Publicly Traded Company. State Owned Enterprise (at time of study; Meridian Energy is partially privatised on 30 September 2013, with 51% shareholding retained by New Zealand Government). Owned by Bosco Connect, retailer, which in turn is owned by Mighty River Power (Gentailer, and State Owned Enterprise at time of study; partially privatised on 5 May 2013). State Owned Enterprise (at time of study; Genesis Energy is partially privatised on 17 April 2014, with 51% shareholding retained by New Zealand Government). Limited Liability Publicly Traded Company, but with 33% ownership by Tauranga Energy Consumer Trust. Retail arm of Mighty River Power (Gentailer, and State Owned Enterprise at time of study; partially privatised on 5 May 2013). Limited Liability Publicly Traded Company.

References Arnot, C., Boxall, P.C., Cash, S.B., 2006. Do ethical consumers care about price? A revealed preference analysis of fair trade coffee purchases. Can. J. Agric. Econ. 54 (4), 555–565. Bénabou, R., Tirole, J., 2006. Incentives and prosocial behavior. Am. Econ. Rev. 96 (5), 1652–1678. Bradley, G., 2009. Switched-on Trustpower makes $105 m. New Zealand Herald, 16 May 2009, pp. C03. Brennan, T.J., 2007. Consumer preference not to choose: methodological and policy implications. Energy Policy 35 (3), 1616–1627. Bushnell, J.B., Mansur, E.T., Saravia, C., 2008. Vertical arrangements, market structure, and competition: an analysis of restructured US electricity markets. Am. Econ. Rev. 91 (1), 237–266. Cameron, A.C., Gelbach, J.B., Miller, D.L., 2011. Robust inference with multiway clustering. J. Bus. Econ. Stat. 29 (2), 238–249. Defeuilley, C., 2009. Retail competition in electricity markets. Energy Policy 37 (2), 377–386. Deller, D., Giulietti, M., Jeon, J.Y., Loomes, G., Moniche, A., Waddams, C., 2014. Measuring consumer inertia in energy, Working Paper. Giulietti, M., Grossi, L., Waterson, M., 2010a. Price transmission in the UK electricity market: was NETA beneficial? Energy Econ. 32 (5). 1165–1174. Giulietti, M., Otero, J., Waterson, M., 2010b. Pricing behaviour under competition in the UK electricity supply industry. Oxf. Econ. Pap. 62 (3), 478–503. Giulietti, M., Waddams, C., Waterson, M., 2005. Consumer choice and industrial policy: a study of UK energy markets. Econ. J. 115 (506), 949–968. Giulietti, M., Waterson, M., Wildenbeest, M., 2014. Estimation of search frictions in the UK electricity market. J. Ind. Ecol. 62 (4), 555–590.

Hortaçsu, A., Madanizadeh, S.A., Puller, S.L., 2015. Power to choose: an analysis of consumer behavior in the Texas retail electricity market. NBER Working Paper. Hortaçsu, A., Puller, S.L., 2008. Understanding strategic bidding in multi-unit auctions: a case study of the Texas electricity spot market. RAND J. Econ. 39 (1), 86–114. Janes, A., 2008. Shareholders let rip at board ‘snouts in trough’. Dominion Post 24 October 2008, pp. 1. Kalayci, K., Potters, J., 2010. Buyer confusion and market prices. Int. J. Ind. Organ. 29 (1), 14–22. Littlechild, S., 2009. Retail competition and electricity markets —expectations, outcomes and economics. Energy Policy 37 (2), 759–763. McFadden, D., 1973. Conditional logit analysis of qualitative choice behavior. In: Zarembka, P. (Ed.), Frontiers in Econometrics. Academic Press, New York, USA, pp. 105–142. Puller, S.L., West, J., 2013. Efficient retail pricing in electricity and natural gas markets. Am. Econ. Rev. 103 (3), 350–355. Sen, S., Gürhan-Canli, Z., Morwitz, V., 2001. Withholding consumption: a social dilemma perspective on consumer boycotts. J. Consum. Res. 28 (3), 399–417. Su, X., 2015. Have customers benefited from electricity retail competition? J. Regul. Econ. 47 (2), 146–182. Waddams Price, C., Zhu, M., 2016. Non-discrimination Clauses: Their Effect on British Retail Energy Prices. Energy J. 37 (2), 111–132. Watt, E., 2008. PM slams Contact directors’ ‘greed’. The Press 23 October 2008, pp. 5. Weir, J., 2008. Contact energy. The Press 27 August 2008, pp. 5. Wilson, C.M., Waddams Price, C., 2010. Do consumers switch to the best supplier? Oxf. Econ. Pap. 62 (4). 647–668.