TL;DR
This paper investigates whether businesses that offer daily deals (like Groupon) are more likely to fail, and if so, whether this is due to the deals themselves or underlying business distress. Using econometric models and Yelp data, they find a significant correlation between unobserved factors driving daily deal adoption and those leading to business failure, particularly in the restaurant sector.
What Problem Does This Paper Address?
The rise of daily deal platforms like Groupon has created a multi-billion dollar industry, but their long-term impact on small businesses remains debated. While some businesses report success, others have described daily deal promotions as detrimental. A key challenge in studying this is isolating the causal effect of daily deals from other factors that might influence both a business's decision to offer a deal and its ultimate survival. This paper aims to address this by exploring whether the unobserved characteristics that lead a business to use daily deals are also correlated with the unobserved factors that contribute to its failure.
Methods
The authors employ econometric techniques to tackle the challenge of unobserved heterogeneity in observational data. They use two primary models:
- Bivariate Probit Model: This model is used to jointly estimate two binary dependent variables: whether a business offered a daily deal and whether it failed. It assumes that the error terms (representing unobserved factors) are jointly normally distributed and allows for cross-equation correlations.
- Seemingly Unrelated Regression (SURE): As a robustness check, they also use SURE, which relaxes the normality assumption of the error terms. This method is particularly useful for analyzing multiple equations with potentially correlated errors.
Both models are used to test for a significant correlation between the unobserved factors influencing daily deal adoption and business survival. They also use AIC (Akaike Information Criteria) for model selection and the Breusch-Pagan test to assess the correlation of error terms in the SURE model.
Datasets
- Groupon Data: A dataset of 16,692 deals offered between January and July 2011 across 20 U.S. cities. The study focuses on a subset of these deals offered in Chicago.
- Yelp Data: A comprehensive dataset of businesses in Chicago, collected from Yelp. This data includes business location, number of reviews, dates of reviews, star ratings, review text, operating status (closed/open), and other features.
The authors created a dataset of 985 businesses that offered daily deals and had Yelp links, referred to as "Groupon-Business." They also compiled a "Yelp Population" dataset of businesses in Chicago, including both those that offered deals and those that did not, to serve as a control group.
Key Insights and Ideas
- Daily Deals as a Distress Signal: The core idea is that a business's decision to offer a daily deal might not be purely strategic but could be influenced by underlying struggles. If these underlying struggles are also linked to failure, then the daily deal itself becomes a signal of potential distress.
- Unobserved Heterogeneity is Crucial: The paper strongly emphasizes that simply regressing failure on daily deal usage is insufficient due to unobserved factors (like management quality, market conditions, or inherent business struggles) that affect both decisions. Econometric models are necessary to account for this.
- Yelp as a Rich Data Source: The study highlights Yelp's utility not just for consumer ratings but as a proxy for business performance indicators like survival, consumer appeal, and revenue. The correlation between Yelp data and actual bankruptcy filings strengthens this claim.
- Restaurants are Particularly Vulnerable: The correlation between unobserved factors and failure is strongest in the restaurant category, suggesting that daily deals might be a more desperate measure for them compared to other businesses like spas.
Key Results
- Significant Correlation: The study found a statistically significant positive correlation between the unobserved factors that influence a business's decision to offer a daily deal and the unobserved factors that contribute to its failure.
- Category Differences: This correlation was strongest in Restaurants & Bars (correlation coefficient of 0.281) and Beauty & Spas (0.246) in the bivariate probit model.
- Model Performance: The joint models (bivariate probit and SURE) performed better than separate univariate models in explaining the data.
- Univariate Model Insights:
- Business Survival: Higher average Yelp ratings and weighted ratings (by number of reviews) were associated with lower failure risk. However, a higher number of reviews (indicating a longer operational history) was associated with increased risk.
- Groupon Adoption: Higher average Yelp ratings and number of reviews were associated with a higher probability of adopting a daily deal. Conversely, higher weighted ratings (rate × nreview) were associated with a lower probability of adopting a deal, suggesting that well-regarded businesses might not need them.
- SURE vs. Bivariate Probit: While the bivariate probit showed stronger correlations, the SURE model provided a useful robustness check, with smaller but still significant correlations.
Limitations and Future Work
- Data Limitations: The study was limited to Chicago and primarily focused on Groupon deals. The authors acknowledge that they did not have enough samples for certain business categories to detect significant correlations.
- Binary Decisions: The models treated daily deal adoption and business failure as binary outcomes. Future work could explore multinomial models for daily deal adoption (e.g., choosing between multiple providers or not offering a deal at all).
- Causal Impact: The current study focuses on correlation. Future work aims to investigate the causal impact of daily deals on metrics beyond survival, such as sales, number of orders, and order size, potentially using instrumental variables.
- Dynamics and Co-integration: The authors plan to investigate the dynamics of daily deal adoption and business failure, testing for stationarity and co-integration to understand long-term equilibrium.
- Broader Data Sources: Future research will incorporate data from other daily deal aggregators (like Yipit) and city revenue departments to expand the scope and potentially include businesses not well-represented on Yelp.
Summary updated