Indicators of hotel profitability: Model selection using Akaike information criteria

Abstract

Research in the important area of evaluating the relationship of hotel room sales with hotel amenities, for instance, food and beverage sales is limited. While this study focuses on differences in overall profitability when considering the transient rate at hotels, the authors have utilized data from hotels with food and beverage operations in order to provide additional insights for this sector. The results show that when modeling the profitability of a hotel, several items are important and should be included when analyzing profitability. The segment of the market, type of hotel, location of the hotel, number of rooms, and average daily rate are all found to be important factors in determining profitability and should be included in models of profit. The transient rate is controlled for in all models but is found to be a significant factor in determining the profitability of a hotel in only some specifications.

Keywords

Hotels model selection Akaike information criteria marketing profitability

Introduction

Research is limited in the crucial area of comparing and contrasting the relationship between hotel room sales and hotel amenities such as food and beverage (F&B) availability. While this study focuses on differences in overall profitability when considering the transient rate at hotels, the authors have utilized data from hotels with F&B operations to provide additional insights for this sector.

Hotels cater to a broad spectrum of clients, each with their own needs, desires, and wants. One natural way to classify different types of guests would be to segment them into two categories: a repeat visitor or renter of rooms, and transient guests, who can be defined as those individuals who stay at a hotel but are not part of a tour booking nor group (Hayes et al., 2011). This distinction is important as the amenities of the hotel may be used in different ways depending upon which category the guest falls into. For example, a guest who is a one-time user may be more likely to use the minibar, while a repeat customer may be more likely to dine in the hotel’s restaurant, if there is one. It is reasonable to question whether these different guest preferences may affect hotel profitability. Researchers or industry analysts often use models that might not be the best application because they do not effectively control for the variables and equations that most closely match the true, underlying relationship.

In examining this potential link, the study asks a broader question of which variables are important in determining the profitability of a hotel and what role the transient rate plays when the Akaike Information Criteria (AIC) is used to measure this sector. Does the average percentage of guests who are transient significantly affect profitability?

The results show that when modeling the profitability of a hotel, several items are important and should be included when analyzing profitability. The segment of the market, type of hotel, location of the hotel, number of rooms, and average daily rate are all found to be important factors in determining profitability and should be included in models of profit. The transient rate is controlled for in all models but is found to be a significant factor in determining the profitability of a hotel in only some specifications.

Food and beverage

F&B has been part of the lodging in the United States since the 1800s when a lodging establishment in Boston, Massachusetts offered a food menu (Levy-Bonvin, 2003). There has been much evolution since then, and the role of hotel F&B in the lodging industry has changed as customer demands progressed.

As trends and the economy vary over time, so do the needs of travelers. F&B operations are meant not just to add to the customer’s experience but are for greater hotel profits. Of course, the economy impacts this. Thus, it is imperative to identify key factors for hotels that rely on profits from F&B so that they can develop a strategic marketing plan based on different travelers’ needs. Focus should be targeted on if F&B revenues do not enhance room revenues and provide profits on their own, then it should be reconsidered or potentially outsourced to a contractor.

Hotel market segments

With occupancy levels rising, hotels are becoming more aggressive, and more strategic, in their pricing. Average Daily room Rates rose 4.2% in 2012 (Mandelbaum, 2013) which is an indication that travelers are increasing their use of hotels. But what segment of travelers is the most beneficial for hotels to target? Quain (1992) identified the benefit of assessing the profitability of different market segments in an effort to increase revenues. In a theoretical situation he describes, the different desires of a transient business traveler compared to a conference traveler making clear the different purchasing patterns between the two. Conference travelers, some of the most transient travelers, tend to be a more captive customer base and partake of hotel F&B where the business traveler is not as much of a consumer as their key need may just be a place to sleep. Vacation travelers, also very transient, on the other hand, tend to have other demands and needs. Typically, business travelers tend to take more advantage of mini bars in hotels where vacationers seek out other options (Cobanoglu et al., 2011).

Banquets tend to provide high revenue compared to the costs and wedding receptions can significantly improve hotel profits while contributing to room revenue, especially transient revenue. The Chicago Hilton Hotel averages over 120 wedding receptions a year which accounts for approximately 10% of total F&B which adds revenue equating to $12 million per year (Adler and Chienm, 2005). Wedding receptions and other special events also represent an important market segment for F&B in hotels. And with weddings comes the potential of more transient guests.

Theoretical model

It is assumed that hotels are profit maximizers who are able to set the price of their rooms. We also allow hotels to discriminate between types of customers: here, transient and repeat guests. There are several reasons for this. Repeat guests may be a member of a frequent guest club that provides for discounts or other rewards. Transient guests may book through discount websites or other medium. It is then reasonable to assume that the price that transient guests pay and the price that repeat guests pay may not be the same. It is assumed, however, that all transients pay the same price and all repeat guests pay the same price. Similarly, the costs associated with both types of guests may differ. Given this, α is the proportion of total guests, Q_a, that are transient, p_t is the room price transients pay, p_r is the price repeat customers pay, w_t is the cost of accommodating a transient guest, and w_r is the cost of accommodating a regular customer. The hotel’s profit function, letting profit equal total revenue minus total cost, is then

π = Q_{a} {α p_{t} + (1 - α) p_{r}} - Q_{a} {α w_{t} + (1 - α) w_{r}}

The quantity

α p_{t} + (1 - α) p_{r}

can be thought of as the weighted average daily rate for any given occupied room. The above equation can be simplified to

π = Q_{a} {α (p_{t} - w_{t}) + (1 - α) (p_{r} - w_{r})}

Profit is then a function of the proportion of transients in the hotel, the average daily rate, and the size of the hotel. In order to determine the effect of the transient rate on profit, the first order condition for the profit function is taken; that is

\frac{δ π}{δ α} = Q_{a} {(p_{t} - w_{t}) - (p_{r} - w_{r})}

It may be assumed that hotels can control the proportion of transient guests; that is, they can choose α to maximize their profit. Setting then the derivative above equal to zero, the equilibrium condition becomes

(p_{t} - w_{t}) = (p_{r} - w_{r})

This implies that, in a market equilibrium, the marginal returns for each type of guest will be the same. Hotels should then set prices and costs in such a way that the above condition is met in order to maximize profit for a given alpha. A natural extension to this study would be to examine how hotels set the different prices for the different guests, but it is not the focus here.

Data

The data used for this research are from hotels across the United States, and in that sense are macroscopic in scope. As such, the results here may not necessarily hold for every hotel, but it does add to the limited literature on what factors influence profitability for hotels with F&B operations. And while there are many characteristics for every hotel that could conceivably impact the profitability of a hotel, it must be noted that it is not possible to control for all aspects of all hotels. Thus, the purpose here is to examine broad, average effects that are felt to impact hotels with F&B.

All of these were hotels with F&B operations and included all hotel regions and type of hotel, as long as they offered F&B. The top five market areas of hotels included in the sample included Los Angeles, New York, Washington, DC, Chicago, and Atlanta. The median number of rooms per hotel was 224.

Methodology

The researchers were able to gain three years’ worth of data from PKF Hospitality Research that was reported by 1242 hotels in the United States for the three calendar years of 2011, 2012, and 2013. While there was not a breakdown between seasons, the data did provide year to year comparisons of the same time periods for all hotels. There are seven segments covered in the data (economy, luxury, midscale—full, upper midscale—full, upper upscale, upscale—full, and upscale—limited), six location types (airport, city center, highway, resort, rural/non-resort, and suburban), and six property types (convention center, convention hotel, full-service hotel, resort hotel, limited-service hotel, suite hotel).

Background on AIC

AIC was first advanced by Akaike (1973) as a way to compare different models on an outcome. An example offered by Snipes and Taylor (2014) discusses that if scholars are interested in what variables influence the ratings of wine and how these variables influence the price of that wine, one might suggest several different regression models. For example, the price of the wine, the type of grape used, or the region the wine was produced in may all play be influential on the rating of a wine. Regression might be run to include price and region information, just price or many other variations. Often, though, the model itself receives only slight thought and is treated as an instrument to reveal show the relationship between the outcome and a specific variable.

The selection of the model is very important, as under-fitting a model may not suggest the true nature of the variability in the outcome variable, while an over-fitted model loses generality. AIC becomes a way to select the best model that balances these downsides. Once the best model is carefully chosen, traditional null-hypothesis testing is then used to define the relationship between specific variables and the outcome of interest.

Candidate models

To examine the potential relationship between transient mix and profitability, and to determine what variables should be included in an analysis of profit, AIC is used. AIC is a method of model selection that selects the best model (in the sense that that model best reflects the true, underlying relationship from among the candidate models) from a group of potential models (Akaike, 1973). Once the best model is selected, more traditional statistical methodologies may be applied. Sixteen candidate models were chosen to examine the relationship between transient mix and profit for each year. The models were then run a fourth time using pooled data from all three years. The results for each of the AIC analysis are then used to determine what variables are important in estimating profitability. The models included in the four timeframes are given in Table 1.

Table 1.

Models estimated for AIC analysis.

M0: profit_i,t = β₀ + ɛ_i,t

M1: profit_i,t = β₀ + β₁(transient_i,t) + ɛ_i,t

M2: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + ɛ_i,t

M3: profit_i,t = β₀ + β₁(transient_i,t) + β₃(adr_i,t) + ɛ_i,t

M4: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M5: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M6: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M7: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M8: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

M9: profit_i,t = β₀ + β₁(transientmix_i,t) + ɛ_i,t

M10: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + ɛ_i,t

M11: profit_i,t = β₀ + β₁(transientmix_i,t) + β₃(adr_i,t) + ɛ_i,t

M12: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M13: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M14: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M15: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M16: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

The variables are given where profit_i,t is the gross profit, transient_i,t is the transient rate, rooms_i,t is the number of rooms in the hotel, adr_i,t is the average rate for the rooms, chainseg_i is the market segment of the hotel, location_i is the general location of the hotel, proptype_i is the type of property, and transientmix_i,t is the transient rate multiplied by the occupancy rate for all locations i in time t. The variable transientmix is meant to capture the average percentage of rooms occupied at a given time by transients as opposed to the percentage of guests who are transient. Including the number of rooms is intended to control for the size of the hotel. PKF defines non-transients as guests who have stayed at the hotel more than once, that is, repeat customers. The transient rate is used in the above specifications as transients could be considered to be more likely to use hotels’ resources and amenities than a non-transient; however, it would also have been valid to use non-transient rates. A subsequent study examining the role of non-transients as opposed to transients would make for an interesting study in its own right. For this study, transients are examined.

The 16 models were selected as they were all felt to be reasonable approximations of the true determinants of profit. They are all linear in nature, in keeping with the theoretical methodology above. AIC is then used to select the best model from the proposed candidates. It should be noted that AIC is not able to select the best model from all possible candidates; that is, it is possible that the model that best reflects the true relationship between profit and transient mix is not in the list above. There could be several reasons for this. For example, the best model may contain variables that were not available in the data or a different mix or permutation of the variables selected may better represent the relationship. Nonetheless, AIC is still very useful as it selects the model that best represents the true relationship from the candidate models for the data given. Kass and Raftery (1995) proposed using the terms “minimal,” “substantial,” “strong,” and “decisive” to correspond to log evidence ratios (LERs) between model probabilities of greater than 0, 0.5, 1, and 2 respectively; we follow their precedent here. R version 3.0.1 was used in the analysis.

Results

AIC results

The AIC statistics are presented below, along with a brief discussion on interpreting AIC statistics. The best model under AIC is the model that has the lowest AIC or AICc score (AICc corrects for sample size). Appendix 1 provides equations for the scores presented as a convenience. Note that when using AIC, the cardinality of the AIC scores is not important; what is important is their ordinal properties. That is, the numbers themselves hold no specific meaning apart from comparing candidate models.

Table 2 presents AIC results for 2013, Table 3 presents results for 2012, Table 4 presents results for 2011, and Table 5 presents results for the pooled data. In this case, model 16 is the best model for year 2013, model 16 for 2012, model 8 for 2011, and model 16 for the pooled data. The AIC weight (w_i) can be thought of as the weight of the evidence of the model with respect to the other models; that is, the model with the highest weight is considered the strongest model with respect to the other models tested. The LER_i is a measure of how decisive the evidence is that the model with the lowest AIC score is in fact the best model. Using the standard stated above, there is decisive evidence that models 8 and 16 are the best models relative to the others. For 2013, 2012, and the pooled data, there is substantial to strong evidence that model 16 is the best model; for 2011, there is substantial evidence that model 8 is the best model. It is then these models whose statistics should be considered; the other models can effectively be ignored.

Table 2.

AIC results for 2013.

Model	df	AIC	AICc	ΔAICc	w_i	ER_i	LER_i
M0	2	43,728.01	43,728.02	1,891.77	0.00	–	–
M1	3	43,712.69	43,712.71	1,876.46	0.00	–	–
M2	4	42,266.78	42,266.81	430.56	2.30e-94	3.13e93	93.50
M3	4	43,468.86	43,468.89	1,632.64	0.00	–	–
M4	5	41,933.91	41,933.95	97.71	4.37e-22	1.65e21	21.21
M5	11	41,908.13	41,908.34	72.09	1.59e-16	4.52e15	15.65
M6	10	41,925.14	41,925.32	89.07	3.27e-20	2.20e19	19.34
M7	10	41,889.03	41,889.21	52.96	2.28e-12	3.16e11	11.50
M8	21	41,837.37	41,837.13	1.88	2.81e-1	2.56	0.41
M9	3	43,722.92	43,722.94	1,886.70	0.00	–	–
M10	4	42,255.68	42,255.71	419.46	5.92e-92	1.22e91	91.08
M11	4	43,482.07	43,482.10	1,645.85	0.00	–	–
M12	5	41,929.87	41,929.92	93.67	3.29e-21	2.19e20	20.34
M13	11	41,905.34	41,905.56	69.31	6.40e-16	1.12e15	15.05
M14	10	41,920.32	41,920.50	84.25	3.65e-19	1.97e18	18.29
M15	10	41,886.17	41,886.35	50.10	9.50e-12	7.57e10	10.88
M16	21	41,835.49	41,835.25	0.00	7.19e-1	1.00	0.00

M0: profit_i,t = β₀ + ɛ_i,t

M1: profit_i,t = β₀ + β₁(transient_i,t) + ɛ_i,t

M2: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + ɛ_i,t

M3: profit_i,t = β₀ + β₁(transient_i,t) + β₃(adr_i,t) + ɛ_i,t

M4: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M5: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M6: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M7: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M8: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

M9: profit_i,t = β₀ + β₁(transientmix_i,t) + ɛ_i,t

M10: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + ɛ_i,t

M11: profit_i,t = β₀ + β₁(transientmix_i,t) + β₃(adr_i,t) + ɛ_i,t

M12: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M13: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M14: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M15: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M16: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

AIC: Akaike information criteria.

Table 3.

AIC results for 2012.

Model	df	AIC	AICc	ΔAICc	w_i	ER_i	LER_i
M0	2	43,515.58	43,515.59	1,957.75	0.00	–	–
M1	3	43,501.48	43,501.50	1,943.66	0.00	–	–
M2		42,000.50	42,000.54	442.70	5.88e-97	1.35e96	96.13
M3	4	43,250.06	43,250.09	1,692.25	0.00	–	–
M4	5	41,651.59	41,651.64	93.80	3.40e-21	2.34e20	20.37
M5	11	41,624.33	41,624.55	66.71	2.60e-15	3.06e14	14.49
M6	10	41,640.25	41,640.43	82.59	9.24e-19	8.60e17	17.93
M7	9	41,613.97	41,614.12	56.28	4.77e-13	1.66e12	12.22
M8	20	41,559.85	41,560.54	2.70	2.06e-1	3.86	0.59
M9	3	43,511.07	43,511.09	1,953.25	0.00	–	–
M10	4	41,988.11	42,988.14	430.30	2.89e-94	2.75e93	93.44
M11	4	43,263.16	43,263.19	1,705.36	0.00	–	–
M12	5	41,646.65	41,646.70	88.86	4.01e-20	1.98e19	19.30
M13	11	41,621.42	41,621.64	63.80	1.11e-14	7.15e13	13.85
M14	10	41,634.33	41,634.51	76.67	1.79e-17	4.45e16	16.65
M15	9	41,610.32	41,610.47	52.63	2.97e-12	2.68e11	11.43
M16	20	41,557.15	41,557.84	0.00	7.94e-1	1.00	0.00

M0: profit_i,t = β₀ + ɛ_i,t

M1: profit_i,t = β₀ + β₁(transient_i,t) + ɛ_i,t

M2: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + ɛ_i,t

M3: profit_i,t = β₀ + β₁(transient_i,t) + β₃(adr_i,t) + ɛ_i,t

M4: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M5: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M6: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M7: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M8: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

M9: profit_i,t = β₀ + β₁(transientmix_i,t) + ɛ_i,t

M10: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + ɛ_i,t

M11: profit_i,t = β₀ + β₁(transientmix_i,t) + β₃(adr_i,t) + ɛ_i,t

M12: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M13: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M14: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M15: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M16: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

AIC: Akaike information criteria.

Table 4.

AIC results for 2011.

Model	df	AIC	AICc	ΔAICc	w_i	ER_i	LER_i
M0	2	43,284.51	43,284.52	1,919.95	0.00	–	–
M1	3	43,286.15	43,286.17	1,921.61	0.00	–	–
M2	4	41,790.81	41,790.84	426.28	2.06e-92	3.68e92	92.57
M3	4	43,036.32	43,036.35	1,671.79	0.00	–	–
M4	5	41,420.69	41,420.74	56.17	4.81e-13	1.58e12	12.2
M5	10	41,401.80	41,401.98	37.41	5.69e-9	1.33e8	8.12
M6	10	41,411.05	41,411.23	46.66	5.58e-11	1.36e10	10.13
M7	9	41,400.83	41,400.98	36.41	9.39e-9	8.07e7	7.91
M8	19	41,363.94	41,364.56	0.00	7.57e-1	1.00	0.00
M9	3	43,283.20	43,283.22	1,918.65	0.00	–	–
M10	4	41,782.00	41,782.03	417.46	1.69e-91	4.48e90	90.65
M11	4	43,043.49	43,043.53	1,678.96	0.00	–	–
M12	5	41,426.62	41,426.67	62.11	2.47e-14	3.07e13	13.49
M13	10	41,403.93	41,404.11	39.54	1.96e-9	3.86e8	8.59
M14	10	41,415.49	41,415.67	51.10	6.06e-12	1.25e11	11.10
M15	9	41,407.24	41,407.39	42.82	3.80e-10	1.99e9	9.30
M16	19	41,366.21	41,366.84	2.28	2.42e-1	3.12	0.49

M0: profit_i,t = β₀ + ɛ_i,t

M1: profit_i,t = β₀ + β₁(transient_i,t) + ɛ_i,t

M2: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + ɛ_i,t

M3: profit_i,t = β₀ + β₁(transient_i,t) + β₃(adr_i,t) + ɛ_i,t

M4: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M5: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M6: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M7: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M8: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

M9: profit_i,t = β₀ + β₁(transientmix_i,t) + ɛ_i,t

M10: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + ɛ_i,t

M11: profit_i,t = β₀ + β₁(transientmix_i,t) + β₃(adr_i,t) + ɛ_i,t

M12: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M13: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M14: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M15: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M16: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

AIC: Akaike information criteria.

Table 5.

AIC results for pooled data.

Model	df	AIC	AICc	ΔAICc	w_i	ER_i	LER_i
M0	2	130,566.8	130,566.8	5,746.24	0.00	–	–
M1	3	130,549.9	130,549.9	5,729.39	0.00	–	–
M2	4	126,158.4	126,158.5	1,337.92	2.53e-291	3.36e290	290.53
M3	4	129,788.3	129,788.3	4,967.79	0.00	–	–
M4	5	125,120.1	125,120.1	299.61	7.43e-66	1.15e65	65.06
M5	11	125,022.4	125,022.4	201.90	1.23e-44	6.93e43	43.84
M6	10	125,073.1	125,073.1	252.58	1.21e-55	7.05e54	54.85
M7	10	125,010.0	125,010.0	189.50	6.04e-42	1.41e41	41.15
M8	21	124,823.8	124,824.0	3.48	1.49e-1	5.71	0.76
M9	3	130,564.9	130,564.9	5,744.41	0.00	–	–
M10	4	126,125.4	126,125.4	1,304.90	3.75e-284	2.27e283	283.36
M11	4	129,821.1	129,821.1	5,000.56	0.00	–	–
M12	5	125,114.2	125,114.2	293.65	1.46e-64	5.83e63	63.77
M13	11	125,017.2	125,017.3	196.73	1.62e-43	5.25e42	42.72
M14	10	125,064.7	125,064.8	244.25	7.79e-54	1.09e53	53.04
M15	10	125,007.5	125,007.5	187.01	2.09e-41	4.06e40	40.61
M16	21	124,820.3	124,820.5	0.00	8.51e-1	1.00	0.00

M0: profit_i,t = β₀ + ɛ_i,t

M1: profit_i,t = β₀ + β₁(transient_i,t) + ɛ_i,t

M2: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + ɛ_i,t

M3: profit_i,t = β₀ + β₁(transient_i,t) + β₃(adr_i,t) + ɛ_i,t

M4: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M5: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M6: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M7: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M8: profit_i,t = β₀ + β₁(transient_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

M9: profit_i,t = β₀ + β₁(transientmix_i,t) + ɛ_i,t

M10: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + ɛ_i,t

M11: profit_i,t = β₀ + β₁(transientmix_i,t) + β₃(adr_i,t) + ɛ_i,t

M12: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + ɛ_i,t

M13: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + ɛ_i,t

M14: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(location_i) + ɛ_i,t

M15: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(proptype_i) + ɛ_i,t

M16: profit_i,t = β₀ + β₁(transientmix_i,t) + β₂(rooms_i,t) + β₃(adr_i,t) + Σ_iβ_i(chainseg_i) + Σ_iβ_i(location_i) + Σ_iβ_i(proptype_i) + ɛ_i,t

AIC: Akaike information criteria.

What is important for this paper, however, is what variables are controlled for in each of the models that are being considered (that is, the models with the lowest AIC scores). For each year, the type of property, the market segment of the property, and location of the property are controlled for. This strongly implies that these controls should be included when modeling the profitability of a hotel. Moreover, in the best models for each year, all three characteristics are included, implying that all three should be included together. All four best models also control for the number of rooms as well as the average daily rate, again implying that these characteristics should be considered.

Regression results

The second purpose of this paper is to examine the potential role the transient rate plays in determining the profit for a hotel. In the four best models, either the transient rate or the transient mix also controlled for. Once the best model is established, the effect of the transient rate or the transient mix can be determined using more standard statistical methodologies. Table 6 presents the regression results for each of the four year combinations.

Table 6.

Regression coefficient estimation results—Profit.

Variable	2013, Model 16	2012, Model 16	2011, Model 8	Pooled, Model 16
transient			−5,798.9 (3,867.7)
transientmix	7,190.6 (5,278.4)	7,959.3** (4,805.1)		5,522.3** (2,886.2)
rooms	38,438.9* (809.9)	34,859.8* (718.5)	30,564.4* (663.4)	34,306.6* (425.8)
adr	37,723.1* (2,195.7)	38,063.4* (2,128.9)	37,397.8* (2,004.2)	38,050.5* (1,226.1)
economy	−3,120,768.8 (2,392,824.3)	−5,236,812.3 (3,484,202.6)		−3,424,579.2** (1,758,160.7)
luxury	−610,108.7 (1,207,099.3)	−845,614.2 (1,082,968.9)	−810,410.2 (997,577.6)	−764,112.8 (638,612.0)
midrange	539,349.4 (3,077,977.7)	222,105.8 (2,751,111.5)	86,514.6 (2,040,395.6)	349,176.5 (1,490,745.3)
uppermidrange	−476,376.0 (1,212,729.8)	−481,257.3 (1,080,943.4)	−652,136.9 (1,005,655.9)	−505,474.9 (641,558.8)
upperupperclass	−1,778,425.2 (1,106,145.6)	−1,634,313.2** (991,807.3)	−1,611,404.2** (917,007.8)	−1,629,283.2* (585,634.1)
upperfull	927,737.6 (1,073,574.2)	780,242.0 (959,546.8)	383,404.3 (904,107.3)	804,489.6 (569,464.0)
convenhotel	−6,452,211* (1,388,568.4)	−5,408,469.0* (1,277,289.9)	−3,697,824.9* (1,202,469.9)	−5,109,411.6* (747,952.7)
fullhotel	399,721.3 (1,123,743.2)	235,093.8 (1,051,109.6)	147,128.36 (997,235.6)	297,047.7 (612,418.7)
limitedhotel	4,652,585.3 (3,747,818.6)			4,023,537 (3,318,668.7)
resorthotel	976,487.1 (1,259,087.9)	650,551.0 (1,173,440.8)	147,667.2 (1,104,296.7)	624,102.2 (683,051.4)
suitehotel	1,408,193.0 (1,213,383.8)	1,276,165.9 (1,129474.4)	1,130,623.5 (1,067,029.7)	1,296,242.8* (658,783)
airport	−749,983.1 (461,879.7)	−699,700.9** (412,728.8)	−511,264 (382,315.1)	−638,245.7* (244,508.3)
citycenter	−1,165,435.4* (396,379.8)	−989,504.7* (353,534.5)	−627,594.5** (327,463.5)	−951,829.0* (209,518.7)
highway	946,408.1 (923,301.3)	683,031.9 (825,514.2)	539,806.6 (761,077.4)	699,840.3 (487,908.6)
resort	−2,237,997.4* (579,725.7)	−2,299,486.5* (524,626.6)	−2,068,068.6* (488,853.0)	−2,252,754.4* (309,738.6)
rural	−485,002.2 (742,928.8)	−367,451.4 (662,657.5)	−202,382.6 (612,581.8)	−372,903.7 (392,425.1)

Numbers in parenthesis are errors.

Variable is significant at 95% confidence.

Variable is significant at 90% confidence.

As shown in Table 6, the transient mix is positive and significant for the pooled data and for 2012, indicating that they are significant predictors of profit. For 2013 and 2011, the various transient rates are insignificant. One thing to note is that in each of the years, the second best model controls for the transient variable not used in the best model; that is, for example, in 2013, the model that includes transient mix is the best model. The second best model is the same, but the transient rate is substituted for the transient mix. Similar results held for the other years as well. Regression results for those second best models were also examined and in each case the various transient rates were found to be insignificant. As the rates are significant in two of the specifications, there is mixed evidence that the transient rate is not a meaningful predictor of profit for this data. As is expected, the coefficient for the average daily rate is positive and significant. It is not directly obvious what the sign for the rooms variable should be, a priori, as the number of rooms is a control for the size of the hotel and not necessarily how many rooms are occupied. As shown in Table 6, however, the coefficient for the number of rooms in the hotel is also positive and significant.

Transient rates and F&B revenue

As the effect of the transient rate is significant in some specifications but not in others, potential effects of the transient rate on sources of revenue were examined; specifically, the methodology above was replicated using the revenue from F&B sales in place of profit. As F&B sales would affect profit, any effect of the transient rate on these sales could be an avenue for the transient rate affecting profit.

In this case, model eight was found to be the best model for all data sets; that is, the best model for all years is

{fbrev}_{i, t} = β_{0} + β_{1} ({transient}_{i, t}) + β_{2} ({rooms}_{i, t}) + β_{3} ({adr}_{i, t}) + \sum_{i} β_{i} ({chainseg}_{i}) + \sum_{i} β_{i} ({location}_{i}) + \sum_{i} β_{i} ({proptype}_{i}) + ɛ_{i}

where the variables are as described above and fbrev_i,t is the revenue from F&B sales for hotel i in time t. Table 7 shows the results from this specification for each of the four years.

Table 7.

Regression coefficient estimation results—Food and beverage sales.

Variable	2013	2012	2011	Pooled
transient	−8,822.12* (3,444.58)	−7,323.35* (3,255.52)	−5,729.92 (3,854.11)	−7,228.70* (1,930.51)
rooms	26,141.12* (734.02)	25,473.34* (679.51)	25,020.18* (661.11)	25,538.91* (397.62)
adr	26,781.95* (1,982.05)	27,213.44* (2,004.79)	28,436.39* (1,997.12)	27,585.99* (1,140.65)
luxury	4,156,436.23* (1,093,949.70)	3,959,777.88* (1,023,720.22)	3,827,617.24* (994,066.5)	3,961,229.17* (596,426.38)
midrange	−337,327.87 (2,786,990.41)	−432,599.96 (2,598,532.54)	71,344.21 (2,033,214)	−162,691.29 (1,390,484.46)
uppermidrange	−1,153,622.95 (1,097,705.75)	−1,074,867.28 (1,020,551.59)	−920,639.24 (1,002,116.47)	−1,050,844.28 (598,428.12)
upperupperclass	−728,541.02 (1,003,452.60)	−703,098.41 (938,540.78)	−582,637.79 (913,780.35)	−675,503.65 (547,489.37)
upperfull	−603,081.09 (972,997.46)	−556,039.55 (907,205.00)	−637,251.51 (900,925.21)	−615,119.98 (532,279.64)
convenhotel	−2,554,641.44* (1,256,548.63)	−2,281,579.08** (1,206,348.16)	−2,837,034.01* (1,198,237.71)	−2,548,401.76* (697,378.35)
fullhotel	−2,754,583.78* (1,016,502.70)	−2,244,303.94* (991,948.06)	−2,825,352.08* (993,725.81)	−2,602,716.32* (570,926.77)
resort hotel	1,097,892.69 (1,139,459.40)	1,339,079.01 (1,107,874.27)	339,641.46 (1,100,410.05)	931,360.19 (636,915.24)
suite hotel	−3,521,221.86* (1,097,434.39)	−2,990,589.13* (1,066,064.18)	−3,538,415.67* (1,063,274.24)	−3,340,826.49* (614,133.83)
airport	−639,745.28 (418,162.96)	−681,481.26** (389,710.37)	−606,155.86 (380,969.51)	−640,825.39 (228,059.93)
city center	−1,812,353.84* (358,648.17)	−1,809,004.30* (333,649.44)	−1,743,327.57* (326,311.00)	−1,788,596.14* (195,296.47)
highway	615,138.58 (834,795.29)	503,924.95 (778,544.97)	684,478.41 (758,398.73)	610,232.05 (454,667.10)
resort	−1,820,304.46* (524,617.85)	−1,830,806.05* (495,317.47)	−1,930,367.96* (487,132.49)	−1,867,405.26* (288,839.90)
rural	−251,444.92 (671,276.56)	−215,909.77 (624,677.07)	−22,584.54 (610,425.78)	−156,141.21 (365,493.71)

Numbers in parenthesis are errors.

Variable is significant at 95% confidence.

Variable is significant at 90% confidence.

As can be seen, the transient rate is negative and significant in three of the four years, indicating that as the transient rate goes up, F&B revenue goes down. This finding has a potential impact on revenue management within the hotel, as transient rates tend to be discounted through group rates and discounted websites, thus lowering revenue even more.

Discussion and future research

The purpose of this paper was twofold. The first was to address what variables or factors are most important in determining the profitability for a hotel. While the factors of hotel profitability have long been established and studied in a multitude of publications, there was a necessity to relook at these areas using AIC with this large sample. This is especially true with the continued growth in the importance of revenue management competencies for hotels. While no academic articles could be identified that utilize AIC in revenue management, Haensel (2012) has begun to demonstrate how AIC could potential be used in hotel revenue management when looking at model fit. The second was to examine the potential link between the percentage of guests who are transient and their effect on profit.

The most important consideration in this paper is that which variables are controlled for in each of the models that are being considered is determined by the models with the lowest AIC scores. And since for each year, the type of property, the market segment of the property, and location of the property are controlled for, it strongly implies that these controls should be included when modeling the profitability of a hotel with F&B operations. Moreover, in the best models for each year, all three characteristics are included, implying that all three should be included together.

Ultimately, this research is successful because it was found that the market segment, hotel type, location, number of rooms, and average daily rate were determined to be the most important factors in determining profit, given the data and models tested. And since all four best models also control for the number of rooms as well as the average daily rate, again implying that these characteristics should be considered, all should be included together when looking at profitability.

Transient guests were found to be a significant predictor of profit in some specifications. This is in keeping with the suggestions from previous research that transient guests, such as conference attendees, tend to utilize F&B options more than regular business travelers (Cobanoglu et al., 2011).

Ultimately, this research reconfirms previous research that finds the importance of specific measurements and their importance to profitability. What the application of AIC does in this study is that it helps reconfirm these measurements through identifying correct model fit. This application is important, as it is the first of its kind to help revenue managers begin to think of new measurements to identify and test against established models of profitability and thus expand the field of revenue management as Haensel (2012) has suggested is possible.

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.

Author Biographies

D Christopher Taylor, PhD is a researcher and wine educator who s expertise is in wine marketing. Currently, he is in the faculty of the University of Houston s Conrad N. Hilton College, where he is also the Director of the Beverage Management Program and the Fred Parks Wine Cellar.

Michael Snipes received his PhD from the University of Colorado - Boulder in 2008. He is currently an Instructor of Economics at the University of South Florida, Sarasota-Manatee. He has conducted research in a number of areas, including family dynamics and the macroeconomic causes of suicide.

Nelson A Barber , PhD, Associate Professor, The Peter T. Paul College of Business and Economics, Hospitality Management Program, University of New Hampshire, Durham, New Hampshire, 03824, 603-862-3303,. Dr. Barber s research has been published in the Journal of Hospitality and Tourism Research, Cornell Hospitality Quarterly, Journal of Marketing Theory and Practice, Journal of Consumer Marketing, among others. His MS is from Purdue University and PhD from Texas Tech University. His research focuses on various dimensions of consumer marketing, pro-environmental consumer behavior, and decision behavior.

Appendix 1. List of equations for AIC statistics

References

Adler

Chienm

(2005) The wedding business: A method to boost food and beverage revenues in hotels. Journal of Foodservice Business Research 7(1): 117–125.

Akaike

(1973) Information theory as an extension of the maximum likelihood principle. In: Petrov

Csaki

(eds) Second International Symposium on Information Theory, Budapest: Akademiai Kiado, pp. 267–281.

Cobanoglu

Berezina

Kasavana

et al. (2011) The impact of technology amenities on hotel guest overall satisfaction. Journal of Quality Assurance in Hospitality & Tourism 12(4): 272–288.

Haensel

(2012) Choice-Set Demand in Revenue Management: Unconstraining, Forecasting and Optimization, Amsterdam, the Netherlands: Alwin Haensel.

Hayes

Ninemeier

Miller

(2011) Foundations of Lodging Management, 2nd edition. Saddle River, NJ: Pearson Education India.

Kass

Raftery

(1995) Bayes factors. Journal of the American Statistical Association 90: 773–795.

Levy-Bonvin

(2003) Hotels–A brief history. Hospitality Net Article, décembre 23(3): 1–8.

Mandelbaum

(2013) US hotel demand powers ahead: Impact on ADR and NOI. Cornell Hospitality Quarterly 52(4): 374–376.

Quain

(1992) Analyzing sales-mix profitability while hotels have used yield management to good advantage to boost revenues, that technique does not consider profitability of various market segments. Profit analysis by segment is a step beyond yield management that can help add dollars to the bottom line. Cornell Hotel and Restaurant Administration Quarterly 33(2): 57–62.

10.

Snipes

Taylor

(2014) Model selection and Akaike information criteria: An example from wine ratings and prices. Wine Economics and Policy 2(3): 3–9.