A Difference-In-Difference Approach to Estimate the Price Effect of Market Entry in High-Speed Rail

Abstract

The effect of competition on prices in the passenger transport sector can be difficult to estimate because of many influencing factors such as state regulation, demand, seasonality, and reactions of indirect competitors. A case of great interest in the rail sector is Italy, where on-track competition in the high-speed segment has been in place for nearly ten years. The paper aims at answering the question whether—and how much—incumbent’s prices are affected when competition starts on a route previously in monopoly. The case is the start of operations on the Turin–Milan–Venice route, where Italo entered in May 2018. Adopting a difference-in-difference approach, we check if Trenitalia changed the price strategy on the route before and after the entry, with respect to two control groups. The addition of placebo tests allows us to understand the ranges of significance of the results and to estimate the noise level of the estimates. Our findings suggest that the start of competition led to lower prices in the short-medium period. In the specific case, the Milano–Venezia route saw Trenitalia’s prices reduced by 21–26% (±2–5%) in a time span of 84–140 days after Italo’s entry and for advanced bookings from 2 to 10 days. Last-day prices remain unchanged, while early bookings are reduced by just 9%. This price reduction does not remain stable in the longer term, when other effects add up and blur the effect of the entry. Nevertheless, we keep observing a smaller, ∼15%, but still negative prices change.

Keywords

rail open-access competition Italy prices difference-in-difference

Introduction

The theoretical expectation regarding the competition effect on a market is a price decline, as monopolists often abuse their dominant position and newcomers must be more efficient and usually price less than the incumbent company. As a consequence, the prices on the liberalized markets are expected to be lower than the counterfactual ones.

In reality, the effect of the competition can be difficult to estimate because of many external influencing factors such as: cost structure, state regulation, demand, seasonality, reactions of indirect competitors, and other factors which can contribute pushing the market prices up or down.

A case of great interest in the transport sector in Italy, where on-track competition in the high-speed rail (and long-distance in general) segment has been in place for nearly ten years. The market is not yet mature and in equilibrium, but the long time since the private competitor Italo started challenging the state-owned company Trenitalia, in addition to its size (about 1/4 of Trenitalia’s entire long-distance offer), let us studying the effects of competition far from early and predatory phases or niche-market cases. However, when studying prices, Beria and Bertolin (2019) observed that, without controlling for other determinants, there was no straight evidence of a lower price level for Trenitalia on routes where Italo is present, with respect to routes where Trenitalia is a monopolist. The reason is that:

[…] competition takes place on the routes with the highest performances in terms of speed and among the main economic areas of the country, both elements that explain potentially higher prices. (Beria & Bertolin, 2019, page 11)

This does not mean that competition has no effect on prices, but that competition takes place where there is room for it in terms of capacity, cost structure, demand, and finally margins. However, the paper is not conclusive whether competition, observing the Italian rail market, is reducing prices as one could reasonably assume.

For this reason, the current paper aims at filling the gap starting from another perspective. Instead of cross-sectional analysis, like in Laroche and Lamatkhanova (2021) derived from the methodology of Fageda and Sansano (2018), the availability of a long observation period before and after competitor’s entry, gives the chance to observe what happens to Trenitalia prices when competition starts. The case is the start of operations on the Turin–Milan–Venice route, previously monopolistic, where Italo entered on May 1^st, 2018. Adopting a difference-in-difference (hereinafter “DID”) approach, we can check if Trenitalia changed the price strategy on the route before and after the entry, with respect to a control group that will be defined later. The DID approach allows controlling over exogenous trends, in particular if Trenitalia has changed prices for the entire network during the period of analysis.

The remainder of the paper is structured as follows. In Section 2, we describe the history of rail market liberalization in Italy. Section 3 reviews the literature on the liberalization effects in rail markets, with a particular focus on price studies in long-distance markets. In Section 4, we introduce the dataset and in Section 5 the methodology is explained. Section 6 starts with descriptive statistics of data and follows with the main model estimates in terms of price reduction effect. Section 7 includes two placebo tests checking the validity of the estimates. Section 8 concludes and discusses the results.

Long-Distance Rail Competition in Italy

As anticipated in the introduction, Italy represents until now¹ a unicum, being the only country with direct on-track competition in high-speed rail services (Fitzová, 2017; Perennes, 2017). Other cases of competition in Europe, such as the well-known cases of the Czech Republic and Slovakia (Tomeš & Jandová, 2018), Sweden (Fröidh & Nelldal, 2015; Vigren, 2017) and Austria (Tomeš & Jandová, 2018; Oszter & Ács, 2021) and more recently Germany (Guihéry, 2020), focus on intercity services or fast conventional services. Overall, the liberalization and governance models in Europe appear to be very heterogeneous (Finger, 2014) and consequently also the outcomes are far from unique.

The story of rail competition in Italy dates back to 2003 when the Decreto Legislativo n.188 of 8 July 2003 early implemented the European Directives on rail competition (2001/12/CE, 2001/13/CE, and 2001/14/CE). According to the decree, any licensed rail company can have access to the national infrastructure manager’s (RFI) rail network and operate both open access or Public Service Obligation (PSO) contracted services (regional and intercity). Interestingly, nothing happened within the PSO segments that remained in-house or assigned after tenders with just one competitor.

The open-access segment, instead, tells a different story. Already in 2006, the first Italian private passenger train company was founded. It was named Arenaways (Boitani & Ramella, 2012; Ramos, 2020) and planned to operate a sort of intercity train between the two northern regions of Piedmont and Lombardy. It started operations later, in 2010, after a number of difficulties (more or less) provoked by the incumbent Trenitalia and its co-owned network operator RFI. To the pitfalls of the incumbent, Arenaways added some unwise operational choices: the fact that its services were quite clearly in competition with PSO fast regional services of Trenitalia made the path more difficult. Arenaways did not last one year with a mutilated service and then declared bankruptcy.

A second company, NTV, was founded in 2007 and started operations in 2012 under the brand Italo (hereinafter in the paper we will use this name). Italo decided to concentrate on the recently opened high-speed tracks between 2006 and 2009 and purchase a large fleet of 25 new high-speed (HS) trainsets. The company accumulated losses until 2015 when thanks to a market repositioning (acknowledging that it cannot be the leading HS company instead of Trenitalia, but a sort of lower cost one) and to the decrease of track access charges to approx. 8€/trainkm made its first profits. Network expansion never stopped and required the purchase of further 26 trainsets until 2021. These further trains run at a maximum speed of 250 km/h instead of >300 km/h of the first group and are used for mixed services (conventional and HS) that today represent part of Italo’s (and Trenitalia’s) networks. In 2018 the company’s majority stocks were sold to an international investor.

Thanks to Italo’s entry, the HS services network in Italy started to evolve yearly, from a previously static situation (Olarte Bacares, 2019) to a continuously increasing number of trainkm (until the COVID-19 crisis). Italo generally did not introduce completely new routes and operate, more or less, the same of Trenitalia, but frequencies and stops of both increased and changed often to follow the demand (Figure 1). Initially, Italo operated on the Torino-Milano-Rome-Naples route (the HS line) but later expanded to further destinations such as Verona, Trento, Bolzano, Udine, etc. (mixed services using fast tracks south of Bologna and conventional ones otherwise). Especially during the summers of the COVID-19 crisis, when demand was less and trainsets unused, Italo (and Trenitalia) started operating also nearly-conventional services to the south of Italy, from Rome to Reggio Calabria. The following Table 1 summarizes the dates of opening of Italo routes.

Figure 1.

Map of Italo services (trains per working day) in 2013 and 2020. Source: our elaborations on official timetables.

Table 1.

Date of Italo’s entries and detail on COVID-19 period changes. Unmentioned lines are operated by Trenitalia only. Source: our elaborations.

Line	Date of Italo’s entry
Main HS line
Milano–Bologna–Firenze–Roma–Napoli	28/04/2012
Extensions to
Torino–(Milano) -…	09/12/2012
Venezia–(Bologna) -…	09/12/2012
Verona–(Bologna) -…	08/12/2015
stops In Ferrara and Rovigo	11/12/2016
Bolzano–(Bologna) -…	01/08/2018
Bergamo–(Bologna) -…	10/03/2019
Udine–(Venezia - Bologna) -…	01/10/2019
Milano–Venezia line
Torino–Milano–Venezia	01/05/2018
Adriatic line
Milano–Rimini–(Ancona)	2013 then closed. Reopened occasionally in summer
COVID-19 period changes
(Roma)–Reggio Calabria	Introduced in summer 2020 and 2021
(Roma)–Bari	Introduced in summer 2021
Bergamo–Roma	Suspended during outbreaks and in summer 2021, then reactivated in Sep 2021
Udine–Roma	Suspended, then reactivated
Trieste–Roma	Summer 2021

It is difficult to estimate Italo’s market share, because of the scarce transparency of both operators on their demand figures, but also because it depends on the perimeter of comparison. Table 2 summarizes the most recent figures of demand available from sparse sources, showing a market share of Italo of at least 20% on the entire long-distance segment.

Table 2.

Estimation of passenger traffic in Italy. Source: our elaborations from CNIT (2018) and company data.

	Billion Pkm (2017)
Trenitalia (long-distance, market)	16.30
Trenitalia (long-distance, PSO)	4.00
Italo (long-distance, market)	5.25
Trenitalia (regional)	18.70
Trenord (regional)	5.40
Other regional companies	3.57
Share Italo on market	24.4%
Share Italo on long-distance	20.5%
Share Italo entire rail market	9.9%

Limiting to HS segment only, figures are scanter and usually come from press releases. According to press, Italo expected to reach 20 Mpax in 2019 (pre-COVID, from 17.5 of 2018), while Trenitalia’s entire market segment is around 40Mpax (Legambiente, 2021). Moreover, Trenitalia operates in monopoly about 100 trains/day under PSO for another 14 Mpax/year (ART, 2017) with some overlaps with market services on some lines. However, it is important to underline that, on pairs where both are present, Italo’s share of supplied trainkm ranges between 30% and 50% or more (Beria & Bertolin, 2019), making it a competitor far from marginal.

Direct competition is, however, not the only relevant aspect characterizing the Italian long-distance rail market (Beria et al., 2018a, 2018b; Beria & Bertolin, 2019; Desmaris & Croccolo, 2018):

a. A peculiar urban geography, with many large and small cities aligned along a limited number of corridors, differently from Spain or France, but also from the network of cities of Germany.

b. A reasonably well-performing rail network, with a few severe saturation problems on some lines, but not on the main corridors Turin–Naples² (thanks to the HS doubling built mainly between 1990 and 2010) or Turin–Venice. Nodes remain instead a problem (today mostly Milan and Florence).

c. A lively coach sector connecting almost every city and a forever-dying national airline (formerly Alitalia³) that opened the doors to the low-cost carriers also on domestic routes, mostly Ryanair, easyJet and recently WizzAir. Overall, competition with air transport is fierce and the capability of each mode to attract passengers from the other depends on users’ characteristics, city pair, and competition level (Capozza, 2016; Bergantino & Madio, 2020).

d. An increasing specialization of rail services by Italo and Trenitalia, accompanied by an excellent quality (onboard services, new rolling stock, etc.).

e. An incredible level of frequency on main long-distance routes, reaching even 5 minutes of headway during peak-hours (Milan–Rome pair was connected 175 times per day in 2020 pre-covid).

f. A relatively scarce integration (especially concerning best cases of Germany or France) between regional and long-distance trains.

Open-Access Competition in Europe and Focus on Prices

European cases

Open-access is one of the possible liberalizations approaches in the very heterogeneous long-distance passenger rail market (Finger, 2014), together with the franchise approach (Wheat et al., 2018) or intermediate forms of duopolies (Montero et al., 2016) and asymmetric regulation (Bougette et al., 2021).

Some theoretical models have tried to show whether head-on competition between a (public) incumbent and a (private) newcomer in the long-distance is financially viable and socio-economically worthwhile. Results are not univocal. Cherbonnier et al. (2017) expect consumers surplus to decrease with competition in the market for most of the simulated conditions (while competition for the market is beneficial). Alvarez-Sanjaime et al. (2015) are slightly less pessimistic, finding that market entries are welfare improving, but only when it generates large traffic increases. Broman & Eliasson (2019) instead affirm that aggregate social welfare increases substantially passing from profit-maximizing monopoly to duopoly competition. In particular, users’ surplus offsets producers’ losses, resulting in stable solutions maximizing social welfare. The reason of welfare losses (aggregate or even users’ one) of open-access, where found, may lay in the possible reduction of economies of scale and network effect of the incumbent, while potential fares reductions and quality increase are likely in most cases (Nash et al., 2019).

Despite the expectations of literature, often pessimistic on the possibility of competition to rise and survive in such a peculiar market, some real-world experiences are available to be studied. Perennes (2017) revises and classifies 34 open-access new entrants in Europe until 2016. These experiences are extremely differentiated in terms of size, business model, industrial organization, and many actually did not survive. Feuerstein et al. (2018) point out the factors mostly influencing the possibility of an open-access competition, including the price level and market potential. Bougette et al. (2021) revise the different barriers to entry and argue that asymmetric regulation and ladder of investment approaches would be beneficial. Seidenglanz et al. (2021) notice that the presence of open-access competition makes incumbents more active also in other markets.

Some of the existing open-access markets have been analyzed, alone or comparatively. The Czech Republic, which shares with Italy the podium of the countries where open-access have deployed more extensively, is characterized by a three-sided competition: the incumbent ČD is challenged since 2011 and 2013 by private operators RegioJet and LeoExpress, respectively (Tomeš et al., 2014 and 2016). Both started from the main line of the country, the Prague–Ostrava, doubling the number of passengers between 2010 and 2016 (Tomeš & Jandová, 2018). The second line of the country, Prague–Brno, opened in 2016 (Tomeš & Fitzová, 2019), reaching also Wien and Bratislava. Much later, newcomers were able also to enter the regional PSO market. Differently from other European countries, the complete vertical separation of Czech railways made the entry easier, but the Czech competitors faced predatory pricing from the subsidized incumbent (Tomeš & Jandová, 2018) and both suffered significant losses. At the same time, also the burden of rail services on public budget rose, both for PSO and discounted fares compensation (Jandová & Paleta, 2019). Slovakia partially shares the history of Czechia, but without “national” competitors, the incumbent was challenged by Czech ones. But, differently from Czechia, the government was openly hostile and introduced numerous advantages for the domestic incumbent, including extensive asymmetrical free tickets programs (Tomeš & Jandová, 2018; Kvizda & Solnička, 2019).

Austria is also open to competition, despite the documented difficulties to get access to essential facilities and useful slots (Tomeš & Jandová, 2018). Differently from Czechia, the newcomer Westbahn chose to concentrate on the main and faster line to Salzburg only. After some years, in 2018, it reached a share of supply almost equal to the OEBB one (Oszter & Ács, 2021).

Sweden was one of the first countries to open and to experience competition (Alexandersson & Rigas, 2013), but for some years the entries were limited to small niche operators (Fröidh & Nelldal, 2015). Fröidh & Byström (2013) underline that in early phases the brand of the incumbent SJ played a significant role in users’ choices. Only in 2015, a larger entrant started operations, the private company MTR, focusing on the main Swedish line to Gothenburg and using fast rolling stock to best exploit the characteristics of the line (Vigren, 2017).

Germany has, similarly to Sweden, a longer history of open-access, but all entries were small niche routes that did not survive for long. Only recently, Flixtrain has started extensive operations of intercity trains in direct competition with DB, but to date we do not have evidence of literature on that (with the notable exception of Guihéry, 2020).

In the UK, open-access competition is limited to routes not served by franchises, but has been effective to stimulate demand. Temple (2015) finds that Grand Central Express stations experienced a larger demand and revenues increase than those where only the franchised operator was present. Similarly, yields on stations with competition increased more slowly than monopolistic ones (+11% vs. +17%). Wheat et al. (2018) look at the cost-side of the open-access model, finding a cost advantage (of 34%) additional to lower input costs. These effects offset the lower economies of density with respect to franchised operators. Using more data, Stead et al. (2019) clarify that open-access operators, in the UK, focus on lower-cost and lower-quality services.

Less known is the case of Poland. Król (2017) introduces the peculiar case of head-on competition between two fully public operators in Poland: from 2009 to 2015 the company Przewozy Regionalne (PR), owned by regional governments, challenged the central government-owned incumbent PKP Intercity (PKP IC). During the period of competition, the effects on passenger figures were clear and the newcomer in competing markets reached a significant market share (33%). The experience ended not only because of the incumbent’s pressures (Król, 2017) but also because of PR industrial and network choices (Król et al., 2018). Król et al. (2019), instead, discuss the post-PR phase, when a number of small and mostly public companies, try to find some place on market niches providing low-performance and low-cost (25–40% cheaper) services.

Limiting to prices and the effect of competition on them, which is the core of this paper, literature is scanter. Again, theory is not straight: on the one side Cherbonnier et al. (2017) affirm that prices decrease is not guaranteed by competition, while Crozet & Chassagne (2013) assume a price decrease at least by the newcomer.

Also, in this case, the real-world observations are useful to settle the question. Laroche and Lamatkhanova (2021) in a cross-routes analysis find that competition has a significant impact on frequencies but not on economy class prices because of duopoly. Tomeš et al. (2016) report of an extensive price war occurred on Czech tracks during the first years of open-access competition. This war led to a sudden reduction of 42% of prices, that together with the doubling of passengers made a +10% overall revenues (Tomeš & Jandová, 2018; Chini et al., 2021). However, no price elasticity is found in the country, whatever is the state of competition (Fitzová et al., 2021). A similar price war occurred in Slovakia, where the incumbent compensated regional trains subtracted passengers and revenues to RegioJet (Kvizda & Solnička, 2019). In Sweden, Fröidh & Nelldal (2015) report the price levels on some routes before the entry of MTR, when only niche operators were on tracks. Vigren (2017) shares the same research question of our work. He estimates the effect of the largest entry in Sweden to date, MTR, finding an average 12% reduction of prices. This value is the average of a concave relationship depending on the number of days in advance the ticket is purchased, where the lowest savings are 31 days before departure and the highest 13 days.

Italy and the effect of competition on prices

Italy, due to its interest, has stimulated a relatively broad literature, with a focus on prices. Early studies found a 30–40% decrease in prices with respect to former Trenitalia prices (see Table 3) in the very first years of entry.

Table 3.

Impact of competition on prices in Italy according to the early studies.

Source	observed price reduction	period	notes
Cascetta and Coppola (2015)	−34%	Between 2011 and 2013	Overall fares reduction
Cascetta and Coppola (2017)	−31%	Between 2009 and 2012	Overall fares reduction
Giuricin (2018)	−40%	Period not specified
NTV (2019)	−30% and −43%	2011–2012 and 2011–2018	Overall yield reduction
Beria et al. (2016)	−15% (observed prices)	Between 2013 and 2014	Milano-Ancona route

Later works looking at a market more in equilibrium find lower reductions. For example, Beria et al. (2019) compare routes with and without competition, finding that Italo is usually 10%–20% cheaper than Trenitalia, but that Trenitalia prices are not significantly different among routes with and without the competitor. The impression is therefore that a step down in Trenitalia prices occurred before 2013, but that later yields remain substantially similar or growing (Beria et al., 2020). Trenitalia remains today the price leader (Bergantino & Capozza, 2018), but the two companies engaged strategic pricing adapting according to rival’s pricing behavior (Bergantino et al., 2015).

Data Description

Our work relies on a large database of Italian train fares collected on a regular basis from the web throughout the years from June 2016 to October 2019. On a selection of 32 origin-destination pairs along the entire national railway network, we collected all the available fares for each train sold by the two main operators, Trenitalia⁴ and Italo/NTV. The simulated purchases have been conducted for various advance booking periods: −20, −10, −5, −2, and −1 days before departure (day 0).

More in detail, for each scheduled train between an origin-destination (OD) pair at pre-defined stations, the following information have been collected:

• Day of purchase and day of travel;

• Train number and train commercial category;

• Station of origin and station of destination;

• Scheduled time of departure and arrival;

• Fare name and level of service;

• Price per fare and level of service (upon availability).

The queries have been conducted on a recurrent basis, both on weekdays and weekends, collecting a sample of 8–31 days per month. It must be noticed that a major issue in surveying happened during April–May 2018, which set the reason for the exclusion of this period from the analysis due to its undersampling. Overall, entire the dataset includes 765 days out of 1125, for a total of slightly less than 1 million single train rides observed for each of the five advance days and including all train types and routes.

Regional trains are excluded from the survey because they are regulated under PSO between the operator and the local public authority, which sets a fixed (and subsidized) price. As well, long-distance Intercity trains are collected, but not considered in the current analysis, being under PSO between the national public authority and the operator, with a capped price⁵. Therefore, only market trains are considered here, for which the operator can set the travel price freely (Table 4).

Table 4.

Train category and regulation. Source: our elaborations.

Train Category	Operator	Regulation	Considered
Frecciarossa	Trenitalia	Market	Yes
Frecciargento	Trenitalia	Market	Yes
Frecciabianca	Trenitalia	Market	Yes
Intercity	Trenitalia	PSO	No
Eurocity	Trenitalia (or partner operator)	Market	No
Thello	Trenitalia	Market	No
Regional	various companies, mostly Trenitalia	PSO	No
Italo	Nuovo Trasporto Viaggiatori (NTV)	Market	(Yes)

Among the whole set of collected fares, the following analysis is focused on the minimum fare available for purchase for each train and for the desired travel date. Consequently, we do not retain the information about the actual flexibility and level of service of the fare the price comes from, assuming that the relevant information for the user is the lowest price to pay to get to his destination.⁶

This paper considers 28 connections out of 32 between origin-destination (OD) pairs in a shorter time frame between May 2017 and May 2019. We use only Trenitalia prices, to observe the incumbent’s behavior in the presence of the newcomer.

These OD pairs have been chosen appropriately to the scope of analysis (see 5.1) and are represented in Figures 2 and 3, according to both the physical layout of the network and the logical grouping made for the analysis. The involved routes cover the whole country and include both high-speed and regular tracks, as well as short, medium, and long travels, to cope with the distance-dependency of unit prices (Beria et al., 2019). We, therefore, calculate the minimum price per kilometer, according to routes length, for each surveyed train. This is the basis for the subsequent analysis.

Figure 2.

Network of surveyed routes and representations of treated OD pairs. Source: our elaborations.

Figure 3.

Representation of control groups’ pairs. Source: our elaborations.

For an overall descriptive analysis of prices, refer to [omissis]. For a quicker overview of the period, routes and train types used here, Table 5 provides the general statistics and the consistency of the sample, approximately half-sized with respect to the original dataset.

Table 5.

Descriptive statistics of the sample used in the analysis. Advanced purchase: −20 days, period: May 2017 to May 2019, Trenitalia market trains only. Source: our elaborations.

relation	group	min	average	max	total obs. Trains (-20 days)
Milano–Brescia	Treated	0.180	0.208	0.434	22,360
Milano–Venezia	Treated	0.058	0.101	0.233	21,637
Torino–Brescia	Treated	0.088	0.125	0.243	7958
Torino–Venezia	Treated	0.050	0.093	0.237	8452
Bari–Ancona	CG1	0.038	0.075	0.178	9454
Bologna–Ancona	CG1	0.045	0.097	0.349	12,909
Bologna–Trieste	CG1	0.067	0.099	0.179	963
Milano–Pisa	CG1	0.033	0.083	0.150	920
Milano–Ancona	CG1	0.039	0.077	0.242	10,075
Milano–Genova	CG1	0.071	0.111	0.207	1500
Milano–Rimini	CG1	0.030	0.087	0.264	10,159
Milano–Udine	CG1	0.055	0.086	0.164	1893
Roma–Bari	CG1	0.040	0.086	0.191	3800
Roma–Genova	CG1	0.040	0.066	0.129	6331
Roma–Reggio C	CG1	0.030	0.081	0.219	3587
Bologna–Firenze	CG2	0.157	0.192	0.589	40,911
Bologna–Venezia	CG2	0.066	0.117	0.298	19,939
Milano–Bologna	CG2	0.046	0.119	0.400	36,947
Milano–Firenze	CG2	0.065	0.110	0.317	17,312
Milano–Napoli	CG2	0.025	0.078	0.204	27,343
Milano–Roma	CG2	0.035	0.095	0.254	39,200
Milano–Torino	CG2	0.104	0.131	0.287	20,030
Roma–Bologna	CG2	0.056	0.097	0.284	54,305
Roma–Firenze	CG2	0.076	0.112	0.264	38,649
Roma–Torino	CG2	0.028	0.082	0.212	13,508
Roma–Venezia	CG2	0.039	0.100	0.272	18,445
Roma–Verona	CG2	0.039	0.091	0.231	7599
Venezia–Firenze	CG2	0.082	0.127	0.354	17,716
					473,902

Problem Conceptualization and Methodology

Setting the problem

If we can observe a sufficiently long period and a sufficiently large sample of individuals, we can test if a variable is changing before and after an event involving a group, with respect to what happens to another comparable unchanged group. In analogy with epidemic studies, we call treated group the first, where something happens, and control group the one where nothing happens.

In the current case, the treated group is represented by OD pairs where competition starts during the observation period. The control group could be a group of comparable OD pairs without competition (monopolistic). However, this control group could be not fully comparable with the treated one, for the reasons already discussed in Beria et al. (2019). For this reason, we perform the analysis also with respect to a second control group of OD pairs where competition was already present before and after the entry event. These routes appear more similar in terms of demand and performances than the residual routes without competition, as we will discuss later in the paper.

Theoretically, one could expect to see a simplified case, as depicted in Figure 4 (a): ticket prices per km of routes in competition are lower than on monopoly ones. The treated group before the event has prices similar to monopolistic ones, while after Italo’s entry they move to a competitive level. In reality, also excluding exogenous effects, the shift could be not “sudden” as in Figure 4 (a), but occur later as in (Figure 4 (b), dashed) because of a “delay” of the incumbent to react. A third case (Figure 4 (b), dotted) is even more likely: incumbent’s prices do not fall suddenly because customers may not be aware or willing to use the newcomer and remain on the incumbent or, conversely, the effect of competition can be anticipated by the incumbent starting to lower price before entry. The competition effect looks then “blurred” in time and may take a while to get to the theoretical equilibrium level of the second control group.

Figure 4.

Theoretical expectations about the effect of competition, without exogenous effects. Source: our elaborations.

Real effect could be even more complex to be seen: exogenous effects may vary all group’s prices and/or controls could be not perfect, as we will discuss in the empirical part.

Our dataset includes a broad number of OD pairs belonging to the three groups schematized in Figure 2. In particular, Italo started operations on the previously monopolistic route (Turin)-Milan-Venice (hereinafter “MV”) on the 1^st of May 2018. The route includes four observed OD pairs: Milano–Brescia (the shortest, 83 km, with significant local demand in addition to long-distance one), Milano–Venezia (the main pair, 258 km), Torino–Brescia (226 km), and Torino–Venezia (the longest, between the two extremes of the line, 401 km). These four constitute our treated group (Table 6).⁷

Table 6.

Analyzed routes and control groups for comparison. Source: our elaborations.

Treated Group (MV Pairs)	Control Group 1, Monopolistic in the Period (CG1 Pairs)	Control Group 2, Already in Competition (CG2 Pairs)
Milano–BresciaMilano–VeneziaTorino–BresciaTorino–Venezia	Bari–AnconaBologna–AnconaBologna–TriesteMilano–PisaMilano–AnconaMilano–GenovaMilano–RiminiMilano–UdineRoma–BariRoma–GenovaRoma–Reggio Calabria	Bologna–FirenzeBologna–VeneziaMilano–BolognaMilano–FirenzeMilano–NapoliMilano–RomaMilano–TorinoRoma–BolognaRoma–FirenzeRoma–TorinoRoma–VeneziaRoma–VeronaVenezia–Firenze

For the comparison, we created two control groups as described above. The CG1 includes OD pairs that remain under Trenitalia monopoly for the entire observation period. None of these pairs includes significant high-speed sections (just a few trains of the Milano–Ancona/Rimini and Roma–Reggio Calabria use part of the HS line, plus the few Milano–Udine running just 84 km on HS until Brescia). Moreover, even if we considered in the sample only market trains, on many of these routes a significant part of the supply belongs to the long-distance PSO contract (Beria & Bertolin, 2019) and this may influence the prices of the market trains considered. This makes this group a probably imperfect control, as the pricing on conventional services may evolve differently than HS or mainline ones.

The second control group, CG2, includes pairs where Italo was already present before and after 01.05.2018. It is more similar to the treated group, being, together with the Milano–Venezia, the core of the Italian train network. It is characterized by high frequencies, high-speed trains only and not or very few PSOs, intramodal competitiveness (airplanes and cars), higher load factors and pairs have generally a fair amount of demand all year round. Of course, every pair has different intermodal characteristics in terms of presence, frequency and competitiveness of other modes, but to have an idea of intermodal competition on considered routes, please refer to Tables 12 and 13 in Annex.

Difference-in-Difference method assumptions.

To analyze the differences in prices of the three groups of city pairs, we create a Difference in Differences (DID, hereinafter) model. DID is a statistical technique used in econometrics and quantitative research studying the differential effect of a treatment between the treated and control group in a natural environment. It calculates the effect of a treatment on an outcome by comparing the average change over time in the outcome variable for the treatment group to the average change over time for the control group.

However, the method has some assumptions (Wing et al., 2018):

• Intervention unrelated to outcome at baseline (allocation of intervention was not determined by outcome)

• Treatment and control groups have common trends in outcome

• Composition of intervention and comparison groups is stable for repeated cross-sectional design

• No spill-over effects.

Real data often do not meet all of these conditions. Therefore, the validity of DID estimates typically revolve around the possible external impulses that displace treatment and control group trends and introduce uncontrolled bias in DID estimates (Besley & Case, 2000; Heckman, 2000; Bertrand et al., 2004; Abadie, 2005). Researchers solve this problem in different ways.

There are matching methods, which select control group members to approximate its characteristics to the treatment group artificially (Somers et al., 2013). Alternatively, a control group is synthesized artificially using weights and then extrapolated to the post-exposure period (Abadie, 2005). In this case, there is a threat of bias due to the artificial transformation of the control group, which ignores possible coordinated shifts in the trends of the control and treatment groups after exposure.

In our analysis, we did not change the original data in order to not lose the information about the actual situation in the passenger transportation market. The same approach is adopted in papers dedicated to the effects of events on rail fares (Li et al., 2020; Wei et al., 2017).

We calculated DID effect of the average monthly price trends of control and treated groups’ routes, which are represented for reference in Figure 5. The DID is calculated symmetrically around the moment of entry of Italo for increasingly longer time spans, ranging from 28⁸ to 336 days. In other words, 28 days means that the DID is between the first 28 days after Italo’s entry and the 28 days before it. The DID is calculated separately for the two control groups.

Figure 5.

Mean monthly available price per km, euro. Source: our elaborations.

Unbiased DID estimates can be obtained if the trends of the control and the test groups are common. In this case, DID estimates do not depend on time. In reality, prices are influenced by a large number of other factors in addition to time, and it is visually impossible to test the hypothesis of common trend assumption for both one factor and multivariate models. To assess whether the inclusion of additional factors is significant in comparison with simple means we compared multivariate model and one-factor model. In addition, we set different intervals (spans) over which the DIDs are evaluated and considered where the condition of parallelism is granted.

Concerning the possible heterogeneity of the control groups, caused by their unique characteristics, we made 20 random subsamples for each control group and obtained 20 DID estimates for each subsample. Thus, we got average DID estimates. By plotting the results, we can see if the whole sample estimates are close to the average values and observe the validity of the control groups.

Finally, to control the possible bias of DID estimates we performed two placebo tests, that will be described in the dedicated Section 7.

The model

The DID algorithm is described in Table 7, where the lower right cell is the DID estimate, or by equation (1), which considers also the control variables. Without including the control variables both methods would give equal results.

Table 7.

The algorithm of obtaining DID estimates with average values only. Source: our elaborations.

Period	MV	CG_i	Difference
After entry	${\bar{l o g P f r}}_{MV, after}$	${\bar{l o g P f r}}_{CGi, after}$	${\bar{l o g P f r}}_{MV, after} - {\bar{l o g P f r}}_{CGi, after}$
Before entry	${\bar{l o g P f r}}_{MV, before}$	${\bar{l o g P f r}}_{CGi, before}$	${\bar{l o g P f r}}_{MV, before} - {\bar{l o g P f r}}_{CGi, before}$
Difference	${\bar{logP f r}}_{MV, after} - {\bar{l o g P f r}}_{MV, before}$	${\bar{l o g P f r}}_{CGi, after} - {\bar{l o g P f r}}_{CGi, before}$	$({\bar{log P f r}}_{MV, after} - {\bar{log P f r}}_{CGi, after}) - ({\bar{log P f r}}_{MV, before} - {\bar{log P f r}}_{CGi, before})$

The effect caused by the entry of Italo on the MV city pair is estimated using the equation

\log (P f r_{r j t}^{i b}) = β_{0}^{i b} + β_{1}^{i b} M V_{r} + β_{2}^{i b} Entr y_{t} + β_{3}^{i b} M V_{r} * Entr y_{t} + X_{t}^{i} β^{i b} + α_{j} + ϵ_{r j t}^{i b}

(1)

where

P f r

is the prices per km of Trenitalia market trains only; r is a selected city pair, I indicates which control group is considered (CG1 or CG2); b is the number of days before departure when train was booked (we observed 1, 2, 5, 10, and 20 days); t is the ordinal number of the day starting from the beginning of observations; X represents exogeneous factors matrix that includes origin, dummies for Frecciabianca and Frecciargento, distance and square distance; peak days and hours (that are Friday pm, Sunday and Monday am), linear time trend;

α_{j}

is the fixed effect cross section dummies and ϵ the residuals. MV stands for Milano–Venezia and represents treated routes dummy. The Entry variable is also a dummy, which points out Italo’s appearance on that line on 01.05.2018.

The subject of the research is the β3 coefficient, which assesses the impact of Italo entry on prices. Since the dependent variable is the logarithm of the price, the value β₃*100 shows the percentage change in price after Italo’s entry.

In our paper, the estimates of the equation parameters (1) are carried out on time spans with a dynamic length. The span is an interval on which the parameters of equation (1) are estimated.

To assess the significance and necessity of including control variables in (1), the parameters of the equation were estimated also with Xβ = 0 in equation (1) and without fixed effects. The β₀ values obtained in this model show the average logarithm of the price of the control group before Italo enters the Milan–Venice route, which is denoted

{\bar{l o g (P f r)}}_{control, before}

. Table 8 systematizes the relationship between beta-coefficients (1) and the average logarithms of prices of the DID method.

Table 8.

Relation between mean values and betas (1) according to DID. Source: our elaborations.

Mean Values	Relation with Beta	Description
${\bar{log (P f r)}}_{control, before}$	$β_{0} (C)$	mean log price for control group before Italo entry
${\bar{log (P f r)}}_{MV, before}$	$β_{0} + β_{1}$	mean log price for MV group before Italo entry
${\bar{log (P f r)}}_{control, after}$	$β_{0} + β_{2}$	mean log price for control group after Italo entry
${\bar{log (P f r)}}_{MV, after}$	$β_{0} + β_{1} + β_{2} + β_{3}$	mean log price for MV group after Italo entry
$({\bar{P f r}}_{MV, after} - {\bar{P f r}}_{CGi, after}) - ({\bar{P f r}}_{MV, before} - {\bar{P f r}}_{CGi, before})$	$β_{3}$	DID estimate

Difference-In-Difference Estimations

Price trends

Before estimating the model, it is useful to visualize (Figure 5) the trends of the average monthly prices for the three groups of routes and for the five advanced bookings (−1 to −20 days to departure).⁹

The absolute level of prices per km, in particular the MV (Milano–Venezia) one, should be considered with caution because it also depends on the distances of the pairs constituting the groups. Although the MV is—all rest equal—an “expensive” route, the price difference is not necessarily the one depicted. However, it is reasonable that the price level of control Group 2 (the routes in competition) is higher than Group 1 (the secondary routes with Trenitalia only) because the former is made of secondary city pairs, less performing trains, slower services and often also by the co-existence of subsidized services.

The trends are, instead, the object of interest of the analysis and of the DID model that follows. The vertical line corresponds to the entry moment of Italo on the MV pairs. Some preliminary comments derive from a simple visual analysis and will be checked by the model.

• It is visible the effect of seasons and holidays (Christmas, Easter, and weekends) which effect can be superimposed on the entry effect, depending on the base of comparison and the period of the comparison. The prices of MV became cheaper if we compare the prices one month before and after the entry, but the prices did not change if we compare prices after competitor’s penetration and a year before (−2 days before the departure).

• Price trends depend on the advanced purchase before the departure. −1 booking day price trends look almost the same, which is expected as Trenitalia stops selling all discounted fares 2 days before departure. “Base tickets” (undiscounted flexible fares), the only available at −1, are basically fixed. On the contrary, the reduction of the price gap with GC2 is well visible in the other cases, especially −5.

• The prices of the CG2 went up after May 2018 especially for late boking. On those routes there were no changes, therefore, Trenitalia could compensate their losses on other competitive routes, or the price changes were caused by other exogenous shifts.

Overall, the series present reasonably parallel trends that make the use of DID feasible. The DID treatment will allow to eliminate the effect of general market trends or exogenous events (e.g., a general repricing of Trenitalia that seem to take place since the beginning of 2019, in correspondence with no entry events). In other words, difference-in-difference method will eliminate inherent price differentials for routes between the treatment and control groups, and time-variant price determinants that are associated with the emergence of a competitor for routes within a treatment group.

β3 coefficients estimations

We perform the estimations of equation (1). DIDs can be calculated as in Table 7 and they coincide with the estimates for the coefficient β3 in equation (1). The estimates with Xβ = 0 and Xβ ≠ 0 do not differ much, but we refer to the controlled ones only, as all the controls are significant.

We can then estimate different β3 coefficients according to the two control groups CG1 or CG2 and assess their sign, level and trend. The results are represented in Figures 6 and 7 with the solid lines.

Figure 6.

Effect of Italo entry (β3 coefficient, representing the percentage reduction in treated prices) with respect to control group CG1 (routes in monopoly), with confidence intervals (±σ). Source: our elaborations.

Figure 7.

Effect of Italo entry (β3 coefficient, representing the percentage reduction in treated prices) with respect to control group CG2 (other routes in competition), with confidence intervals (±σ). Source: our elaborations.

The values are not always constant in time, which means that the effect of the entry of Italo on the Milano–Venice route depends on the length of the span during which the average is calculated. Since this may violate the DID hypotheses, we must perform some checks.

First, we tested the validity of the control groups. We made 20 random subsamples and calculated DID estimates for each of the advanced booking and control groups. It let us calculate the average DID estimates and the confidence intervals for them, represented in Figures 6 and 7 with dashed lines. The whole sample DID estimates are close to the average simulated DID estimates which shows the validity of the control groups.

Second, we considered only the situations where DID estimates are constant, which is needed for the DID estimates to be correct. In the CG1 graphs, there are no stable DID estimates intervals, which confirms it as an imperfect control group, as expected. Concerning the CG2 graph (and ignoring for a while the results for −1 days), we can recognize three phases of all trends:

1. From 28 to 84 days, the effect falls from a positive result to a minimum. Even if on the one side we do not expect a sudden step on prices before and after the entry (Italo is less known on the route, many Trenitalia tickets were already booked, Trenitalia does not need to correct instantaneously prices but can wait and see what happens), the non-constancy of the betas indicates a potential violation of the hypothesis of DID method.

2. From 84 to 140 days spans, the DID is constant and with negative sign, especially for CG2. It means that in this range the effect of Italo’s entry is correctly measured. The negative sign means that the entry of Italo has reduced the prices of Trenitalia. We will comment later on the values of such reduction.

3. For spans longer than 140 days, the difference does not remain stable but grows asymptotically, sign that the effect of the entry is averaged by successive price changes. However, for CG2, the effect of Italo remains negative, while the differences year-on-year versus CG1 reach the 0% level (end of the effect).

The trend of the −1 days advanced booking, is totally different in sign and slope. This is coherent with the fact, already commented, that one day before departure only the fully-flexible “base price” is available. Being “base price” fixed on a route and not depending on demand, it is coherent to find no effect in the DID.

Table 9 summarizes the β3 values found in the reliable time span of 84–140 after Italo’s entry. Negative signs mean reduction. The value is the percentage changes of Trenitalia’s prices due to Italo’s entry.

Table 9.

Effect of Italo entry (β3) for both control groups calculated by equation (1) with Xβ≠0 symmetrical and lagged spans. Source: our elaborations.

Booking days	DID Symmetrical Spans
	β3 CG1			β3 CG2
	84	112	140	84	112	140
1	0.02	0.04	0.07	0.00	−0.01	−0.02
2	−0.08	−0.11	−0.16	−0.21	−0.22	−0.21
5	−0.06	−0.10	−0.16	−0.26	−0.26	−0.25
10	−0.07	−0.18	−0.20	−0.25	−0.23	−0.23
20	0.11	−0.06	−0.10	−0.19	−0.17	−0.17

To verify that the DID estimates versus CG2 are stable as they look in the middle intervals (84–140 days), we tested the equality of means by ANOVA, whose results are shown in Table 10. Again, CG1 confirms to be unsuitable as a control, since all cases show a p-value smaller than 5%. On the contrary, all estimations versus CG2, but −1 day, are significant with p-values well above 5%.

Table 10.

The test of the hypotheses about the equality of the means. Significant values in bold. Source: our elaborations.

Booking day	Span (Phase 2)	DID Symmetrical Spans
		CG1		CG2
		Mean DID	p-value ANOVA	Mean DID	p-value ANOVA
−1	84–140	0.07	0.00	−0.01	0.00
−2	84–140	−0.15	0.00	−0.21	0.73
−5	84–140	−0.15	0.02	−0.26	0.71
−10	84–140	−0.18	0.00	−0.24	1.00
−20	84–140	0.07	0.02	−0.18	0.25

In summary, Trenitalia always reduced prices on Milano–Venezia route with respect to both control groups. In the ranges of validity (CG2 and 84–140 days after Italo’s entry), we observe a reduction of 21–26% of Trenitalia prices for advanced bookings from 2 to 10 days. The effect on earlier bookings, −20 days, is slightly smaller, at 17–19%, a fact already found by Vigren (2017) in Sweden. As already said, the price one day before does not change.

Having in mind also Figure 5, the mean prices on CG1 routes went down, while Trenitalia mostly raised prices on competitive ones CG2: Trenitalia does not overprice on monopolistic routes, because they are the “poorer” in terms of demand, performance and willingness to pay of users. On the contrary, it was able to keep higher prices on the “richer” competitive routes. This fact apparently contradicts with competition theory. However, on the Milano–Venezia, the entry of Italo generates a steep reduction of prices in the medium and long run, which is probably attributable to the increased total capacity provided by the two players and just maybe to a more aggressive price attitude of the former monopolist.

Placebo Tests

To test the significance of the DID estimates we developed two placebo tests. The first shows the differences in the two control groups (CG2 routes against CG1). The second is used to determine how accurate are the estimations.

Placebo test 1

In the first placebo, we consider CG2 routes as treated and CG1 as a control group. The idea is that if the two controls were “perfect,” the DID between them would give 0.

The test is realized by equation (2)

\log (P f r_{r j t}) = β_{0} + β_{1} C_{2 r} + β_{2} Entr y_{t} + β_{3} C_{2 r} * Entr y_{t} + X_{t} β + α_{j} + ϵ_{r j t}

(2)

where

C_{2} = 1

for CG2 routes and

C_{2} = 0

for CG1 routes.

The test shows that the 2 control groups behave differently. Their differences are around zero just at the beginning of the observation period (28–84 days). Onwards, the two groups start to diverge significantly reaching a difference of 15–20% on a yearly basis. This quantifies what we already found in the qualitative analysis in Section 6.1: the prices on routes in competition (CG2) went up compared to the monopolistic ones (CG1) since August 2018. This is the sign of a change of strategy of Trenitalia for richer HS routes that was not applied to the other non-HS routes belonging to the CG1. Such event is independent of the entry of Italo on the Milano–Venezia treated route (it is later and for the whole country) and does not invalidate the comparison. (Figure 8)

Figure 8.

First Placebo test. CG2 is compared to CG1. Source: our elaborations.

Placebo test 2

The second placebo is more important because can be used to define the accuracy of the estimates. We randomly divided the city pairs constituting each of the control groups in two subgroups each. The first subgroup was considered as a control, and the second was “treated” on 05/01/2018. This breakdown was done several times and the error of the DID estimates coming from the control groups were estimated. If each control group is homogeneously behaving in the time period, we expect that the DID estimates are close to zero.

Figure 9 reports the test for the two control groups. First of all, it is clear that CG2 is more homogenous, with DID estimates ±5%: all routes/pairs included are served by similar train categories (HS or mixed HS), similar speed, similar demand density (main cities), etc. This level can be considered a margin of error. The test was performed 20 times (random subgroups made) and we obtained the same level of the noise in the model.

Figure 9.

Second Placebo test. Average DID estimates within each control group. Source: our elaborations.

Figure 10 represents the standard deviation of the DID estimates of the control groups for each booking day. It represents the error of the average estimate and can be used to correct the value found for the effect of Italo’s entry on the route. Limiting to the range of time spans and control group where the estimation can be considered reliable (84–140 days and CG2), the error values found are very limited (less than 3% except for −5 days, with less than 5%). Out of these boundaries (less than 84 days or above 140 days and vs. CG1), the error is more significant confirming the unreliability of those results.

Figure 10.

Error estimation of control groups (DID estimates standard deviations of placebo tests). Source: our elaborations.

Finally, Table 11 represents the final DID estimates considering the standard deviations values of the effect of Italo’s entry into the market. The bold values are those significant according to a double test (significantly different from zero according to the Student’s test (5%) plus ANOVA test represents the mean values homogeneity of the DID estimates >5%).

Table 11.

Final DID estimates considering standard deviation values in 84–140 days span. Source: our elaborations.

Booking days	DID Symmetrical Spans
	CG1		CG2
	DID	p-val ANOVA	DID	p-val ANOVA
−1	0.07±0.01	0.004	−0.01±0.01	0.000
−2	−0.1T5±0.04	0.004	−0.21±0.02	0.732
−5	−0.15±0.03	0.020	−0.26±0.05	0.710
−10	−0.18±0.05	0.000	−0.24±0.03	0.997
−20	0.07±0.02	0.022	−0.18±0.02	0.245

Conclusion and Discussion

Estimating the price effect of competition is a complex task, because:

a. Price is not the only lever of competition;

b. The effect measured on a short time span (e.g., two specific days) can be deeply biased by the presence of exogenous uncontrolled factors;

c. A simple before-and-after estimation does not consider the situation in the whole market;

d. It depends upon what is chosen as a reference.

In this paper we solved most of the issues listed above. We adopted a Difference-in-Difference approach to eliminate the problem of exogenous changes. At the same time, we have a very long observation period (2 years of nearly continuous price observations), during which some OD pairs saw the entry of a second competitor in addition to the incumbent. We also have a pool of different routes to build control groups and test their reliability.

The treated route is the Torino–Milano–Venice route, in Northern Italy, where the private train company Italo entered in a market previously occupied by the national company Trenitalia. The route is a mixed HS-conventional line, quite fast and with a significant demand density. We observed the prices of four OD pairs on the route, of different length. The period of analysis is one year before and one after the entry date, 1 May 2018. We also built two control groups. CG1 is made of routes in monopoly for the entire period. Various tests prove that this group is an imperfect control because of its heterogeneity, but also because it is made of services too different from the nearly-HS one of the treated routes. Almost all untreated routes, in fact, are not fast, connect secondary markets and are often overlapped with PSO intercity services with capped prices. The second control group, CG2, proved to be much better. It is made of city pairs belonging to the mainline connecting the North with Rome and Naples.

We performed the estimations, obtaining the average DID effect of Italo’s entry on Trenitalia’s prices. The effect is a percentage and represents how cheaper has become Trenitalia’s average prices over periods from 28 to 336 days after entry, with respect to the corresponding symmetrical periods before entry.

We observe 3 main phases, especially with reference to CG2:

1. A transition period of about three months during which the average of Trenitalia available prices reduces thanks to the extra-capacity, plus the lower prices of Italo.

2. A stable period from about 3 to about 6 months, that can be used to see the actual and unbiased effect of Italo’s entry. The observed price effect is a reduction of Trenitalia’s available prices of 21–26% for advanced bookings between 2 and 10 days. The effect on earlier bookings is smaller, about 18%, 20 days before departure. The effect on prices of the day before departure is nearly zero, since Trenitalia did not change its flexible fares (the only available 24hrs before departure). Overall, we can affirm that the extra-capacity provided by Italo impacted more positively on general and business travelers who book later, than leisure ones that tend to book earlier.

3. A long-term effect, during which the entry of Italo is blurred with other exogenous events and price changes of treated and control routes. The long-term effect is smaller, but still negative, of about −15% prices reduction.

The paper and its results however present some limits, which affect the quality of the assessments: the violation of the common trend assumption and the heterogeneity of the control groups. To cope with these limits, we have restricted the ranges of validity and performed some placebo tests. The parallelism of trends is guaranteed if the DID-estimates are stable in time, which happens only in the period 84–140 days and CG2, also according to the ANOVA test. This is the only case when the estimate is representing only the entry effect. As the DID estimates can be biased by the heterogeneity of the control groups, we performed a second placebo test (a randomly chosen pairs in a control group vs. the whole group) which let us obtain the standard deviation of the estimates. The values represent the noise, whose level is as low as 2–5% for the stable period and CG2, and is used to correct the initial DID estimates.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Paolo Beria

Notes

Appendix

Table 12.

Supply levels of the OD pairs analyzed (long-distance only). Source: our elaborations from companies’ timetables.

OD pair	Average Rail supply [trains/day per direction]^*			Air supply [flights per day]^**	Coach supply [coaches per day]^**
	Trenitalia	Of which PSO	NTV
Bari–Ancona	15	5	—	—	3
Bologna–Ancona	20	6	—	—	7
Bologna–Bolzano	6	0	—	—	5
Bologna–Firenze	43	1	20	—	27
Bologna–Trieste	3	2	—	—	4
Bologna–Venezia	20	0	8	—	11
Milano–Pisa	6	5	—	—	2
Milano–Ancona	13	2	—	—	9
Milano–Bologna	41	5	10	—	26
Milano–Brescia	26	0	—	—	—
Milano–Firenze	19	1	10	—	18
Milano–Genova	12	11	—	—	6
Milano–Napoli	28	2	15	14	5
Milano–Rimini	13	2	—	—	5
Milano–Roma	39	0	17	34	19
Milano–Torino	20	0	12	—	22
Milano–Udine	2	0	—	2	3
Milano–Venezia	26	0	—	—	10
Roma–Bari	4	1	—	7	14
Roma–Bologna	57	2	20	4	18
Roma–Ferrara	5	2	2	4	2
Roma–Firenze	40	0	20	—	29
Roma–Genova	9	2	—	6	10
Roma–Reggio C	7	4	-	6	5
Roma–Torino	13	0	12	9	7
Roma–Venezia	21	2	8	6	4
Roma–Verona	8	0	4	4	6
Torino–Brescia	10	0	—	—	2
Torino–Venezia	10	0	—	1	3
Venezia–Firenze	18	2	8	—	6

^*Average number of train/day per direction based on the current offer in a sample of days in 2017.

^**Number of flight or coach/day per direction based on the supply of Wednesday, 31^st of October 2018.

Table 13.

Distance and travel time and for OD pairs analyzed. Source: our elaborations from companies’ timetables.

OD Pair	Distance [km)	Rail		Air	Coach
		HS Trains	Conventional
Bari–Ancona	442	3h40'	4h10'	—	6h30′ –7h30′
Bologna–Ancona	218	1h50'	2h0'	—	3h00′ –3h30′
Bologna–Bolzano	261	2h30'	2h40'	—	4h00′
Bologna–Firenze	95	0h40'	1h10'	—	1h15′ –1h30′
Bologna–Trieste	296	3h0'	3h50'	—	5h30′
Bologna–Venezia	151	1h10'	1h30'	—	2h00′ –2h30′
Milano–Pisa	301	—	4h10'	—	4h30′ –5h15′
Milano–Ancona	429	3h0'	4h10'	—	5h30′ –6h15′
Milano–Bologna	213	1h10'	2h20'	—	3h00′ –3h30′
Milano–Brescia	83	0h40'	0h50'	—	—
Milano–Firenze	306	1h50'	3h50'	—	4h00′ –5h30′
Milano–Genova	140	1h30'	1h40'	—	2h00′ –2h30′
Milano–Napoli	790	4h30'	8h50'	1h20′	10h00′ –13h00′
Milano–Rimini	330	2h10'	3h20'	—	4h45′ –5h15′
Milano–Roma	567	3h10'	6h50'	1h10′	8h00′ –10h30′
Milano–Torino	143	0h50'	1h30'	—	2h00′ –2h30′
Milano–Udine	365	4h0'	4h0'	0h55′	5h30′
Milano–Venezia	258	2h10'	2h20'	-	3h30′ –4h00′
Roma–Bari	498	4h0'	6h20'	1h05′	5h30′ –6h30′
Roma–Bologna	356	2h10'	4h10'	0h55′	5h00′ –6h00′
Roma–Ferrara	400	2h40'	4h40'	0h55′	5h00′ –5h30′
Roma–Firenze	261	1h30'	2h50'	-	3h30′ –4h00′
Roma–Genova	494	4h0'	5h0'	1h 10′	6h30′ –8h30′
Roma–Reggio C	663	4h50'	7h10'	1h 10′	9h00′ –10h00′
Roma–Torino	711	4h10'	—	1h 15′	10h00′ –10h30′
Roma–Venezia	504	3h30'	5h40'	1h 05′	6h30′ –7h00′
Roma–Verona	507	3h0'	—	1h 00′	7h00′ –7h30′
Torino–Brescia	226	1h40'	2h30'	—	3h15′
Torino–Venezia	401	3h10'	4h10'	1h 25′	6h00′ –6h30′
Venezia–Firenze	243	1h50′	—	—	3h30′ –4h00′

References

Abadie

(2005) Semiparametric difference-in-differences estimators. The Review of Economic Studies, 72(1), 1–19. https://doi.org/10.1111/0034-6527.00321

Alexandersson

Rigas

(2013). Rail liberalisation in Sweden. Policy development in a European context. Research in Transportation Business & Management, 6, 88–98. https://doi.org/10.1016/j.rtbm.2012.12.004

Álvarez-SanJaime

Ó.

Cantos-Sanchez

Moner-Colonques

Sempere-Monerris

J. J.

(2015) A model of internal and external competition in a High Speed Rail line. Economics of Transportation, 4(3), 178–187.

ART (2017) Relazione sugli esiti dell’indagine conoscitiva sull’analisi dei profili regolatori inerenti il mercato dei servizi di trasporto via autobus a media e a lunga distanza in regime di libera concorrenza. Autorità di Regolazione dei Trasporti.

Bergantino

A. S.

Capozza

(2018). Pricing strategies: who leads and who follows in the air and rail passenger markets in Italy. Appl. Econ, 50(46), 4937–4953. https://doi.org/10.1080/00036846.2018.1459039

Bergantino

A. S.

Capozza

Capurso

(2015). The impact of open access on intra-and inter-modal rail competition. A National Level Analysis in Italy. Transport Policy, 39, 77–86. https://doi.org/10.1016/j.tranpol.2015.01.008

Bergantino

A. S.

Madio

(2020). Intermodal competition and substitution. HSR versus air transport: Understanding the socio-economic determinants of modal choice. Research in Transportation Economics, 79, 100823. https://doi.org/10.1016/j.retrec.2020.100823

Beria

Bertolin

(2019). Evolving Long-distance passenger services. Market concentration, fares and specialisation patterns in Italy. Res. Transp. Econ, 74, 77–92. https://doi.org/10.1016/j.retrec.2019.01.004

Beria

Grimaldi

Albalate

Bel

(2018a). Delusions of success: Costs and demand of high-speed rail in Italy and Spain. Transport Policy, 68, 63–79. https://doi.org/10.1016/j.tranpol.2018.03.011

10.

Beria

Nistri

Laurino

(2018b). Intercity coach liberalisation in Italy: Fares determinants in an evolving market. Research in Transportation Economics, 69, 260–269. https://doi.org/10.1016/j.retrec.2018.07.029

11.

Beria

Redondi

Malighetti

(2016) The effect of open access competition on average rail prices. The case of Milan–Ancona. Journal of Rail Transport Planning & Management, 6(3), 271–283. https://doi.org/10.1016/j.jrtpm.2016.09.001

12.

Beria

Tolentino

Bertolin

Filippini

(2019). Long-distance rail prices in a competitive market. Evidence from head-on competition in Italy. Journal of Rail Transport Planning & Management, 12, 100144. https://doi.org/10.1016/j.jrtpm.2019.100144

13.

Beria

Tolentino

Filippini

(2020) Are prices reduced from direct competition in high-speed rail? Some unexpected evidences from Italy. MPRA Paper No. 98841. https://mpra.ub.uni-muenchen.de/98841/

14.

Bertrand

Duflo

Mullainathan

(2004) How much should we trust differences-in-differences estimates?. The Quarterly Journal of Economics, 119(1), 249–275. https://doi.org/10.1162/003355304772839588

15.

Besley

Case

(2000) Unnatural experiments? Estimating the incidence of endogenous policies. The Economic Journal, 110(467), 672–694. https://doi.org/10.1111/1468-0297.00578

16.

Boitani

Ramella

(2012) Arenaways e altre storie ferroviarie (poco edificanti). Mercato concorrenza regole, 14(1), 97–116.

17.

Bougette

Gautier

Marty

(2021) Which access to which assets for an effective liberalization of the rail sector. Competition and Regulation in Network Industries, 22(2), 87–110. https://doi.org/10.1177/17835917211012326

18.

Broman

Eliasson

(2019) Welfare effects of open access competition on railway markets. Transportation Research Part A: Policy and Practice, 129(1), 72–91. https://doi.org/10.1016/j.tra.2019.07.005

19.

Capozza

(2016). The effect of rail travel time on airline fares: First evidence from the Italian passenger market. Economics of Transportation, 6, 18–24. https://doi.org/10.1016/j.ecotra.2016.06.002

20.

Cascetta

Coppola

(2015) New high-speed rail lines and market competition: Short-term effects on services and demand in Italy. Transportation Research Record, 2475(1), 8–15. https://doi.org/10.3141/2475-02

21.

Cascetta

Coppola

(2017) Evidence from the Italian high-speed rail market: competition between modes and between HSR operators. In High-Speed Rail and Sustainability. Routledge, pp. 82–95.

22.

Cherbonnier

Ivaldi

Muller-Vibes

Van Der Straeten

(2017) Competition for versus in the market of long-distance passenger rail services. Review of Network Economics, 16(2), 203–238. https://doi.org/10.1515/rne-2018-0007

23.

Chini

Kvizda

Solnička

Špetík

(2021). Assessment of market power on the route Prague–Ostrava. Transportation Research Procedia, 55, 252–259. https://doi.org/10.1016/j.trpro.2021.06.029

24.

CNIT (2018) Conto Nazionale delle Infrastrutture e dei Trasporti - anni 2017-2018. Istituto Poligrafico e Zecca dello Stato SA.

25.

Crozet

Chassagne

(2013) Rail access charges in France: Beyond the opposition between competition and financing. Research in Transportation Economics, 39(1), 247–254. https://doi.org/10.1016/j.retrec.2012.06.021

26.

Desmaris

Croccolo

(2018). The HSR competition in Italy: How are the regulatory design and practices concerned? Research in Transportation Economics, 69, 290–299. https://doi.org/10.1016/j.retrec.2018.05.004

27.

Fageda

Sansano

(2018). Factors influencing prices and frequencies in the interurban bus market: Evidence from Europe. Transportation Research Part A, 111, 266–276. https://doi.org/10.1016/j.tra.2018.03.022

28.

Feuerstein

Busacker

(2018). Factors influencing open access competition in the European long-distance passenger rail transport—A Delphi study. Research in Transportation Economics, 69, 300–309. https://doi.org/10.1016/j.retrec.2018.03.001

29.

Finger

(2014). Governance of competition and performance in European railways: An analysis of five cases. Utilities Policy, 31, 278–288. https://doi.org/10.1016/j.jup.2014.03.003

30.

Fitzová

(2017) European railway reforms and efficiency: Review of evidence in the literature. Review of Economic Perspectives, 17(2), 103–120. https://doi.org/10.1515/revecp-2017-0005

31.

Fitzová

Kališ

Pařil

Kasa

(2021) Competition in long distance transport: Impacts on prices, frequencies, and demand in the Czech Republic. Research in Transportation Business & Management, p. 100655.

32.

Fröidh

Byström

(2013). Competition on the tracks–Passengers’ response to deregulation of interregional rail services. Transportation Research Part A: Policy and Practice, 56, 1–10. https://doi.org/10.1016/j.tra.2013.09.001

33.

Fröidh

Nelldal

B. L.

(2015) The impact of market opening on the supply of interregional train services. Journal of Transport Geography, 46(3), 189–200. https://doi.org/10.1016/j.jtrangeo.2015.06.017

34.

Giuricin

(2018). Ownership change heralds expansion at Italo-NTV (article at Railway Gazette International). Railway Gazette, (5), 1–4.

35.

Guihéry

(2020). Autocars interurbains en Allemagne: pour FlixBus, le temps des interrogations. TI&M, Transports, Infrastructures & Mobilité, 521, 33.

36.

Heckman

J. J.

(2000) Causal parameters and policy analysis in economics: A twentieth century retrospective. The Quarterly Journal of Economics, 115(1), 45–97. https://doi.org/10.1162/003355300554674

37.

Jandová

Paleta

(2019) Impact of on-track competition on public finances–The case of the Czech Republic. Journal of Rail Transport Planning & Management, 12(3), 100145. https://doi.org/10.1016/j.jrtpm.2019.100145

38.

Król

(2017) Open access competition in the long-distance passenger rail services in Poland. Yearbook of Antitrust and Regulatory Studies (YARS), 10(16), 155–166.

39.

Król

Taczanowski

Jarecki

Kołoś

(2019) Publicly-owned operators can also challenge incumbents. New cases of open-access passenger rail competition in Poland. Journal of Rail Transport Planning & Management, 12(1), 100150. https://doi.org/10.1016/j.jrtpm.2019.100150

40.

Król

Taczanowski

Kołoś

(2018). The rise and fall of Interregio. Extensive open-access passenger rail competition in Poland. Research in Transportation Economics, 72, 37–48. https://doi.org/10.1016/j.retrec.2018.06.008

41.

Kvizda

Solnička

(2019) Open access passenger rail competition in Slovakia–experience from the Bratislava–Košice line. Journal of Rail Transport Planning & Management, 12(3), 100143. https://doi.org/10.1016/j.jrtpm.2019.100143

42.

Laroche

Lamatkhanova

(2021) Effects of open access competition on prices and frequencies on the interurban railway market: Evidence from Europe. Research in Transportation Business & Management, p. 100705. https://doi.org/10.1016/j.rtbm.2021.100705

43.

Legambiente (2021) Rapporto pendolaria 2021.

44.

Wang

Zhang

(2020). Are conventional train passengers underserved after entry of high-speed rail?-Evidence from Chinese intercity markets. Transport Policy, 95, 1–9. https://doi.org/10.1016/j.tranpol.2020.05.017

45.

Montero

J. J.

Ramos Melero

(2020) Competitive ten dering for rail track capacity: The liberalization of railway services in Spain. EUI Working papers. RSCAS 2020/27.

46.

Montero

J. J.

Ramos

Giuricin

(2016) Open with care: the duopoly model for the transition to competition in long-distance passenger railway transportation. Competition and Regulation in Network Industries, 17(3–4), 241–259. https://doi.org/10.1177/178359171601700303

47.

Nash

Smith

Crozet

Link

Nilsson

J. E.

(2019). How to liberalise rail passenger services? Lessons from european experience. Transport Policy, 79, 11–20. https://doi.org/10.1016/j.tranpol.2019.03.011.

48.

NTV (2019) Italo - nuovo trasporto viaggiatori. Company profile november 2019. NTV.

49.

Olarte Bacares

C. B.

(2019) Influence of the evolution of high-speed railway infrastructure on the success of Italian liberalization. Compet. Regul. Netw. Ind, 20(2), 113–137. https://doi.org/10.1177/1783591719847615

50.

Oszter

Ács

(2021) When market opening reverses – comparative analysis of public transport policy of multiple Central and Eastern European countries. European Transport Conference.

51.

Perennes

(2017). Open access for rail passenger services in Europe: Lesson learnt from forerunner countries. Transportation Research Procedia, 25, 358–367. https://doi.org/10.1016/j.trpro.2017.05.413

52.

Ramos

(2020) Competition in the market of rail passenger services: the experience of small and medium-sized new entrants in long-distance services in Europe. In Handbook on Railway Regulation. Edward Elgar Publishing. https://doi.org/10.4337/9781789901788.00026

53.

Seidenglanz

Taczanowski

Król

Horňák

Nigrin

(2021). Quo vadis, international long-distance railway services? Evidence from Central Europe. Journal of Transport Geography, 92, 102998. https://doi.org/10.1016/j.jtrangeo.2021.102998

54.

Somers

M. A.

Zhu

Jacob

Bloom

(2013) The validity and precision of the comparative interrupted time series design and the difference-in-difference design in educational EvaluationMDRC working paper on research methodology.

55.

Stead

A. D.

Wheat

Smith

A. S.

Ojeda-Cabral

(2019) Competition for and in the passenger rail market: Comparing open access versus franchised train operators' costs and reliability in Britain. Journal of Rail Transport Planning & Management, 12(1), 100142. https://doi.org/10.1016/j.jrtpm.2019.100142

56.

Temple

(2015). Open access long distance passenger rail services in the United Kingdom: The grand central experience. Transportation Research Procedia, 8, 114–124. https://doi.org/10.1016/j.trpro.2015.06.047.

57.

Tomeš

Fitzová

(2019). Does the incumbent have an advantage in open access passenger rail competition? A case study on the Prague–Brno line. Journal of Rail Transport Planning & Management, 12, 1–11. https://doi.org/10.1016/j.jrtpm.2019.100140

58.

Tomeš

Jandová

(2018) Open access passenger rail services in Central Europe. Research in Transportation Economics, 72(1), 74–81. https://doi.org/10.1016/j.retrec.2018.10.002

59.

Tomeš

Kvizda

Jandová

Rederer

(2016) Open access passenger rail competition in the Czech Republic. Transport Policy, 47(1), 203–211. https://doi.org/10.1016/j.tranpol.2016.02.003

60.

Tomeš

Kvizda

Nigrin

Seidenglanz

(2014) Competition in the railway passenger market in the Czech Republic. Research in Transportation Economics, 48(1), 270–276. https://doi.org/10.1016/j.retrec.2014.09.052

61.

Vigren

(2017). Competition in Swedish passenger railway: Entry in an open access market and its effect on prices. Economics of Transportation, 11–12, 49–59. https://doi.org/10.1016/j.ecotra.2017.10.005

62.

Wei

Chen

Zhang

(2017) Demand shocks, airline pricing, and high-speed rail substitution: Evidence from the Chinese market. Journal of Transport Economics and Policy, 51(4), 266–289. https://doi.org/10.2139/ssrn.3030930

63.

Wheat

Smith

A. S.

Rasmussen

(2018). Can competition for and in the market co-exist in terms of delivering cost efficient services? Evidence from open access train operators and their franchised counterparts in Britain. Transportation Research Part A: Policy and Practice, 113, 114–124. https://doi.org/10.1016/j.tra.2018.03.004

64.

Wing

Simon

Bello-Gomez

R. A.

(2018). Designing difference in difference studies: best practices for public health policy research. Annual Review of Public Health, 39, 453–469. https://doi.org/10.1146/annurev-publhealth-040617-013507