Intervention (Hidden)
Our intervention will specifically focus on restaurants in the city of Rome. The location choice for this project is guided by the necessity for finding sufficient and comparable restaurants within easy reach, while at the same time being able to differentiate between sectors differing in their baseline levels of formality. Within Italian cities, we expect Rome to offer some of the highest levels of heterogeneity regarding formality levels prevalent in different parts of the city. The businesses active in central, affluent, and touristed parts of the city are for example different in this regard from those active in more remote and less prosperous parts of the city.
The product our design focuses on is a standard dining option for 8-12 people. We chose this because the costs are sufficiently high to warrant a detailed interaction between the restaurant manager and the customer. Additionally, there is considerable scope for evasion opportunities, and restaurants usually exhibit a bit of leeway in how to price such gatherings.
To conduct the survey, we plan to send enumerators to engage in “mystery shopping” and elicit price quotes for a hypothetical dinner party. These interactions are intended to leave restaurant staff convinced they just witnessed a routine interaction with a potential customer. The key aspect of our survey will be the random variation in how the price elicitation will be carried out. Before entering a given establishment, we will randomize between two receipt-related “cues”: debit card (control group) and cash (treatment group). In the control group, enumerators will ask the price of the meal stating that they will pay with debit card. In the “cash” treatment group, the mystery shopper will state that they will pay in cash. While restaurants in Italy are liable for VAT at 10% on their sales, the two methods of payments offer very different evasion opportunities.
Due to the full randomization of the establishments into the control and treatment groups, a simple comparison of the outcomes should already provide a well-identified estimate of the effect of these cues on price-setting behavior. However, since we expect to find substantial variation in prices across restaurants, we believe including restaurant-level fixed effects will substantially improve our statistical power. Therefore, each restaurant will be interviewed on two separate occasions, by distinct enumerators. One interview will be in the control group, while the other in the treatment group. We plan to randomize the order in which these two visits occur to ensure we can test whether there is a systematic difference in prices quoted for the first versus the second visit. In addition to improving statistical power, this has the advantage of allowing us to directly quantify how evasion rents are passed through to the consumer at each restaurant, permitting, and thus to study the distribution of evasion rent-sharing.
In our pre-intervention data collection, we have collected several characteristics on 1200 restaurants in three urban areas of the city of Rome, including size (i.e., establishment assets and employees), rating (based on Google reviews), activity in food delivery platforms, and legal status (limited vs. unlimited liability). This allows us to estimate heterogeneity in treatment effects depending on these observables. Specifically, we are interested in whether evasion rent sharing differs depending on high- vs low-formality (proxied by urban area, activity on food delivery platforms, and legal form of the business).In addition, we randomly vary the size of the planned dinner party (8 versus 12 guests) in order to gauge the effect the size of the proposed transaction has on the observed price difference. These additional features of our survey will be completely cross-randomized into the original treatment structure. Therefore, we can estimate all main effects even when disregarding the additional layers of heterogeneity. Finally, we will be able to observe enumerator demographics, including age and gender. Even though we will not be able to cross-randomize allows us to study whether demographic characteristics of potential customers (e.g. gender or age) influence the degree of rent-sharing. Unfortunately we will not be able to control which enumerators will goto each individual restaurant, meaning this aspect of the survey will not be mechanically cross-randomized with other aspects of the survey.
Finally, restaurants will be visited a third time by a different enumerator. These visits will differ from the main visits of the survey as enumerators will not conduct an interview. Instead, they will ask to conduct a small transaction (e.g. having a small lunch or an aperitivo) and record whether a legally valid receipt is issued. This will allow us to study another, very important margin of heterogeneity, i.e. whether treatment effects differ between high- and low-evasion-risk restaurants. The estimated VAT gap in Italy is amongst the highest in the EU, and since restaurants are small firms one might reasonably expect many of the cash transactions would be hidden from tax authorities. However, these additional visits will allow us to more accurately measure evasion risk. One should note that on the other hand, given the third party-reported nature of digital transactions, the probability for the restaurant owner to evade the transaction is effectively zero in the control group.