Hotels’ occupancy rates and convergence: Empirical evidence from the first pandemic wave

Abstract

The scope of this study is to unravel possible convergence clubs regarding the occupancy rates of the hotel industry in the US states in the aftermath of the first COVID-19 pandemic wave. For this reason, the underlying paper resorts to the application of the generic convergence algorithm developed in Phillips and Sul (2007) for a sample of 50 US states over the daily period ranging from 01.12.2019 to 26.07.2020. The empirical analysis supports the identification of two primary convergence clubs consisting of an equally distributed number of regions (states). However, the two clubs can be merged into one after the implementation of the Phillips and Sul (2009) methodology, revealing that the first pandemic wave has eliminated any distinct (economic) characteristic between the different US states.

JEL codes: L10, L80, R10 .

Keywords

occupancy rates club convergence COVID-19 transition paths US states

Introduction

Since the dawn of 2020, the global hotel industry has been hit hard by the COVID-19 disease see among others, (Assaf and Scuderi, 2020). The coronavirus (SARS-CoV-2), which was first reported in China back in December 2019, has been declared a pandemic by the World Health Organization on the 11^th of March 2020. This pandemic crisis has posed unprecedented health, social, and economic restraints, affecting every sector of the economy including the hotel industry. The present ongoing crisis represents one of the greatest challenges for the hotel industry. The response to such a challenge is not an easy task for policymakers and stakeholders (hotel managers, market participants, other related professionals, etc.) due to the dynamic “profile” of the pandemic.

Several studies analyze the effect of the pandemic crisis on the performance of the hotel industry (Boto-García and Leoni, 2021; Kock et al., 2020; Lee et al., 2020; Payne et al., 2021; Tsionas, 2020) by employing several econometric techniques. However, the investigation of convergence/divergence patterns in the hotel industry amid the pandemic crisis has been nearly overlooked by the existing studies since most of them focus solely on the convergence of tourism demand (Mérida et al., 2016). To draw sharp inferences concerning the convergence behavior of the US hotel industry and offer some primary policy strategies considering the gradual recovery of the sector, this study lends support to the application of the Phillips and Sul (P-S) algorithm to identify several convergence clubs (clusters) among the spatial units of the US sample.

The relevant study has several novelties. First, and foremost, this is the first attempt to our knowledge’s best for the investigation of possible convergence clusters in the US hotel industry during the first pandemic wave. Second, the econometric analysis relies on a high-frequency aggregated panel dataset consisting of daily data to better capture the related volatility of the sample variables compared with most of the existing studies that use monthly or even annual data. Third, the convergence analysis employs the occupancy rate as the key influential factor that directly reflects consumers' demand and hotel revenues (Yeon et al., 2020) compared to most of the related studies that rely on prices and hotel revenues to quantify hotel performance.

Data description and methodological framework

Sample and variables

We utilize a sample of hotel occupancy rates (OCC) at a U.S. state level over the first pandemic wave period. The sample consists of a balanced daily panel dataset comprising 50 US states over a 9-month period starting from 01.12.2019 to 26.07.2020 yielding 11,950 observations (N = 50 and T = 239). The sample variable of this study is obtained by the Smith Travel Research (STR) hotel database which tracks global hotel supply and demand and is widely used by related studies (see for example Assaf et al., 2020; Assaf and Tsionas, 2018; Assaf et al., 2017; Hall and Parker, 2020).¹ The sample covers the first pandemic wave that hit the US states during the first months of 2020 until the end of the summer. The sample reports the average occupancy rate from all the available hotel categories (luxury, midscale, upper midscale, upper upscale, and upscale) per U.S. state.

As it is evident, from the occupancy rates, there is a non-linear “N-shaped” curvature trend over the sample period in nearly all 50 US states (see Figure 1). This finding is mainly attributed to the dramatic changes in airline schedules and operations during March 2020 and April 2020. According to the Bureau of Transportation Statistics, there was a 96% annual passenger decline from April 2019 and a 51% decline from March 2019. However, a closer look at the density plots of the occupancy rates uncovers some differences between the US states (see Figure 2). Specifically, for some states (see, e.g.,, Alaska, Arkansas, Connecticut, Delaware, Illinois, Indiana, Idaho, Michigan, Mississippi, Oklahoma, Oregon, Tennessee, and West Virginia), a bi-modal distribution is revealed. On the contrary, a multi-modal distribution of the OCC density plots is traced for the following states: Alabama, Arizona, California, Massachusetts, New York, and North Dakota.

Figure 1.

Occupancy rate trends (%).

Figure 2.

Density plots of the occupancy rate.

Methodology

To analyze the convergence rates of OCC at the US state level, we apply the Phillips and Sul (2007, 2009) methodology. The relevant analysis consists of two stages. First, we define a single factor model as

v_{i, t} = β_{i, t} λ_{t}

(1)

where

β_{i}

indicates the different distance among the systematic part of

v_{i, t}

and

λ_{t}

. Then, the behavior of

β_{i, t}

can be expressed as

β_{i, t} = β_{i} + σ_{i} ϑ_{i, t} L {(t)}^{- 1} t^{- α}

(2)

where

ϑ_{i, t}

iid (0,1)

across

i

, whereas the term

β_{i}

is fixed. Also, the term

L (t)

indicates a varying function having

L (t) \to \infty

t \to \infty

, whereas

β_{i, t}

convergences to

β_{i}

for all

α \geq 0

. We can then decompose

v_{i, t}

v_{i, t} = f_{i t} + α_{i t}

(3)

v_{i, t} = (\frac{f_{i t} + α_{i t}}{λ_{t}}) λ_{t} = β_{i, t} λ_{t} for all i and t

(4)

According to Phillips and Sul (2007), the transition coefficient can be expressed as

η_{i t} = \frac{v_{i, t}}{1 / N \sum_{i = 1}^{N} v_{i, t}} = \frac{β_{i, t}}{1 / N \sum_{i = 1}^{N} β_{i, t}}

(5)

Where

β_{i, t}

represents the transition coefficient and

η_{i t}

is the relative transition parameter, which converges to unity. The cross-sectional variance of

β_{i, t} (σ_{t}^{2})

can then pre-presented as

σ_{t}^{2} = \frac{1}{N \sum_{i = 1}^{N} {(η_{i t} - 1)}^{2}} \to 0 as t \to 0

(6)

The $t$ test of the null hypothesis of convergence implies that: $ℋ_{0} : β_{i} = β and α \geq 0$ , whereas $ℋ_{1} : β_{i} \neq β for all i or α < 0$ .

We can then define the cross-sectional variance ratio $Δ_{1} / Δ_{t}$ , as

Δ_{t} = \frac{1}{N \sum_{i = 1}^{N} {(η_{i t} - 1)}^{2}}, η_{i t} = \frac{v_{i, t}}{1 / N \sum_{i = 1}^{N} v_{i, t}}

(7)

Finally, the $t$ statistic $(t_{\hat{b}})$ for $\hat{b}$ can be estimated as

\log (Δ_{1} / Δ_{t}) - 2 log L (t) = \hat{α} + \hat{b} log t + {\hat{u}}_{t}, for t = [ρ Τ], [ρ Τ] + 1, T with ρ > 0

(8)

In the regression below, we follow Phillips and Sul (2007), having $L (t) = log (t + 1)$ , $\hat{b} = 2 \hat{α}$ and $ρ = 0.3$ . We can reject the null hypothesis of convergence at a 5% level, if $t_{\hat{b}} < - 1.65$ , having $η_{i t} \to 1$ and $Δ_{t} \to 0$ as $t \to \infty$ .

Results and discussion

Table 1 reports the empirical findings of the club clustering algorithm based on the P-S methodology (Phillips and Sul, 2007). The analysis proceeds to two steps. At the first stage, we test the validity of the null hypothesis of convergence for the whole sample, while at the second stage, we continue with the delineation of possible separate convergence clubs.

Table 1.

Results from club convergence classification.

Phillips and Sul (2007)				Phillips and Sul (2009)
Category	log t	t-stat	New club	Final classification	log t	t-stat
Full sample	0.270	51.931	-	-	-	-
Club 1 [AL, AK, AZ, CA, CO, DE, GA, ID
IN, MD, MI, MS, MT, NV, NC, OR, SC, SD
TN, TX, UT, VA, WY]	0.952 [0.476]	11.264	1+2	Club I [0.135]	0.270	51.931
Club 2 [AR, CT, FL, HI, IL, IA, KS, KY, LA, ME, MA, MN, MO, NE, NH, NJ, NM, NY, ND, OH, OK, PA, RI, VT, WA, DC, WV, WI]	0.529 [0.264]	72.187

Note: The critical value is −1.65. The number in brackets denotes the speed of convergence (α) for each club cluster.

The results of the primary P-S algorithm do not indicate rejection of the null hypothesis (convergence) for the full sample since the estimated t-statistic (51.931) is greater than the critical value of −1.65. This implies the existence of full convergence for the whole sample. However, we must check for the existence of different club clusters since based on Phillips and Sul (2007, p. 1798) “[…] Examples include the possible existence of convergence clusters around separate points of equilibria or steady-state growth paths.” From our analysis, we have identified two convergence clubs. Specifically, Club one consists of Alabama, Arkansas, Arizona, California, Colorado, Denver, Georgia, Idaho, Indiana, Maryland, Michigan, Mississippi, Montana, Nevada, North Carolina, Oregon, South Carolina, South Dakota, Tennessee, Texas, Utah, Virginia, and Wyoming, while Club two includes the rest 28 US states. As in the case of the full sample, the P-S algorithm fails to reject the null hypothesis of convergence for the two initial formulated clubs since the estimated t-statistics are greater than the critical value.

Having delineated the primary convergence clubs based on P-S (2007) generic algorithm, the analysis proceeds with the interpretation of the speed of convergence (α) among the formulated clubs.² Our findings suggest that the speed of convergence is positive in both clusters, though its magnitude varies across the two formulated clusters. Specifically, the first club records an absolute value of α = 0.476, indicating a higher adjustment speed to convergence than the other club (α = 0.264). It is noteworthy that this value is almost two times greater than the one that appears in Club 2. This means that the 28 US states formulating Club 2 are approaching one another more slowly in relative terms compared to the rest US states, uncovering a different demand pattern.

Our findings when utilizing Phillips and Sul (2009) methodology uncover that we fail to reject the null hypothesis of convergence in two cases (Club 1 and Club 2), revealing that the two initial convergence clubs can be merged into one covering the full sample (50 US states). We notice that the newly formulated club (Club I) has a low estimated convergence speed equal to α = 0.135. This result is in alignment with our previous findings, unraveling a slow path to full convergence.

Finally, Figure 3 illustrates the transition path of the two identified convergence clubs under the assumption of the sample’s overall convergence to unity. The transition curves displayed reveal whether the identified clubs converge or diverge from below or above unity over the examined period. Our findings suggest that the transition path of Club one consisting of 22 states, exhibits a convergent behavior approaching cross-sectional average (unity) from above. On the contrary, the other club including the rest of 28 US states (Club 2) follows a different convergent pattern approaching cross-sectional average (unity) from below. From the shape of the relevant curvatures, it is evident that the transition paths confirm the convergence test results.

Figure 3.

Transition paths based on the P-S (2007) algorithm.

Conclusions

The present study applies the convergence methodological framework developed by Phillips and Sul (2007, 2009) to assess for the first time in the empirical literature the convergence patterns of occupancy rates in the 50 US states over the first pandemic wave. By utilizing daily data over the period December 2019 to July 2020, the empirical evidence suggests that the null hypothesis of full convergence for the whole sample cannot be rejected. Moreover, we have identified two initial convergence clubs with a relatively moderate convergence speed, alongside their transition paths that can be merged into one after the implementation of the Phillips and Sul (2009) methodology. The results clearly illustrate the existence of a convergence pattern in the hotel occupancy rates throughout the COVID-19 first wave.

The empirical findings unfold how different methodological approaches can be applied by policymakers and hoteliers to trace the different consumers’ demand patterns of the hotel industry and better fine-tune their policies/strategies. The existence of a fully convergent behavior approaching cross-sectional average (unity) for both initial clusters (Club one and Club 2) indicates that the first pandemic wave has eliminated any distinct (economic) characteristic between the different US states (natural beauty and destination characteristics, travel promotion, branding, state support, regional income, number of international arrivals, etc.) distorting to a large extent the performance of the US hotel industry. The results from the convergence analysis support the study by Plzáková and Smeral (2021) emphasizing the dramatic decrease in tourism demand due to containment measures. Plzáková and Smeral (2021) assert that restrictions in transportation and mobility, alongside the closure of hotels and restaurants can initiate in the post-pandemic period price increases or quality declines on tourism services. In addition, due to the economic crisis initiated by the pandemic and the containment measures, there is a strong possibility that the tourism industry will not be able to reach breakeven occupancy levels at pre-crisis prices. Such an outcome encompasses useful (horizontal) strategies to hoteliers and market participants to better adjust their services and facilities provided, considering the gradual reopening of the industry in the aftermath of the pandemic crisis. This in turn will provide the policymakers with the ability to target and respond more efficiently to the dynamic nature of the COVID-19.

One of the most vital managerial strategies that should be implemented to mitigate the negative impact of the COVID-19 pandemic crisis is not only to rely on the necessary infrastructure measures including inter alia cleaning, sanitizing, and air handling systems but to invest in more flexible technologies such as artificial intelligence, which will allow contactless processes and mobile hospitality. The future challenge for managers and policymakers of the tourism industry is the identification and the adjustment to the new segments of demand emerged by the COVID-19 crisis.

Footnotes

Acknowledgments

The authors would like to thank Professor Albert Assaf (the Editor) and an anonymous referee of this journal for fruitful comments and suggestions that enhanced the merit of the paper. All errors belong to the authors. The usual disclaimer applies.

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 iDs

Michael L Polemis

Panayiotis Tzeremes

Nickolaos G Tzeremes

Notes

Author biographies

Michael Polemis is Associate Professor of Industrial Economics at University of Piraeus and a Member of the Hellenic Competition Commission. His main areas of research include industrial organization, competition policy, regulatory economics, energy, and environmental economics. He has published more than 95 scientific papers in journals such as the European Journal of Operational Research, Economics Letters, Regional Studies, The Energy Journal, Energy Economics, International Business Review, Journal of Productivity Analysis and Journal of Regulatory Economics, among others.

Panayiotis Tzeremes is Assistant Professor at the Department of Accounting and Finance at University of Thessaly. His main research interests are in applied microeconomics analysis and economic analysis. He has published in international journals such as Journal of Corporate Finance, European Journal of Operational Research and Economics Letters.

Nickolaos Tzeremes is a Professor of Microeconomic Analysis at University of Thessaly. Currently he serves as an Associate Editor of Economic Modeling. His work has appeared in high-ranked journals such as the Research Policy, European Journal of Operational Research, International Journal of Operations and Production Management, Economics Letters, OMEGA, Environment and Planning A, Journal of the Operational Research Society, among others.

References

Assaf

Scuderi

(2020) COVID-19 and the recovery of the tourism industry. Tourism Economics 26: 731–733.

Assaf

Tsionas

(2018) The estimation and decomposition of tourism productivity. Tourism Management 65: 131–142.

Assaf

Josiassen

Woo

Agbola

F.W

Tsionas

(2017) Destination characteristics that drive hotel performance: a state-of-the-art global analysis. Tourism Management 60: 270–279.

Assaf

Atkinson

Tsionas

(2020) Endogeneity in multiple output production: evidence from the US hotel industry. Tourism Management 80: 104124.

Boto-García

Leoni

(2021) Exposure to COVID-19 and travel intentions: evidence from Spain. Tourism Economics 135481662199655. DOI: 10.1177/1354816621996554.

Hall

Parker

(2020) Why go home? pearl jam’s ‘home shows’ & hotel demand. Annals of Tourism Research Empirical Insights 1: 100001.

Kock

Nørfelt

Josiassen

, et al. (2020) Understanding the COVID-19 tourist psyche: the evolutionary tourism paradigm. Annals of Tourism Research 85: 103053.

Lee

C-C

Chen

M-P

(2020) The impact of COVID-19 on the travel and leisure industry returns: some international evidence. Tourism Economics 135481662097198. DOI: 10.1177/1354816620971981.

Mérida

Carmona

Congregado

, et al. (2016) Exploring the regional distribution of tourism and the extent to which there is convergence. Tourism Management 57: 225–233.

10.

Payne

Gil-Alana

Mervar

(2021) Persistence in Croatian tourism: the impact of COVID-19. Tourism Economics 135481662199996. DOI: 10.1177/1354816621999969.

11.

Phillips

PCB

Sul

(2007) Transition modeling and econometric convergence tests. Econometrica 75: 1771–1855.

12.

Phillips

PCB

Sul

(2009) Economic transition and growth. Journal of Applied Econometrics 24: 1153–1185.

13.

Plzáková

Smeral

(2021) Impact of the COVID-19 crisis on European tourism. Tourism Economics 135481662110311. DOI: 10.1177/13548166211031113.

14.

Tsionas

(2020) COVID-19 and gradual adjustment in the tourism, hospitality, and related industries. Tourism Economics 135481662093303. DOI: 10.1177/1354816620933039.

15.

Yeon

Song

Lee

(2020) Impact of short-term rental regulation on hotel industry: a difference-in-differences approach. Annals of Tourism Research 83: 102939.