A discrete choice hotel room demand model

Abstract

An embodiment models the demand and pricing of hotel rooms. The embodiment receives historical data for multiple previous guests and generates a multinomial logit ("MNL") model with demand shock variables, which are represented using MNL utility parameters. The embodiment estimates the MNL utility parameters using likelihood maximization and determines the demand shock parameters using the estimated MNL utility parameters. The embodiment then forecasts future demand for hotel rooms based on the demand shock parameters.

Description

[Technical Field]

[0001] Field One embodiment is directed generally to computer systems, and more particularly to computer systems that generate hotel room demand models. [Background technology]

[0002] Background information Revenue management is the process of dynamically adjusting the price of a product or service in response to changing market conditions or shifts in supply. The revenue management process was pioneered by the passenger airline industry and has been imitated by other industries, including cargo airlines, hotels, car rentals, shippers, and advertising intermediaries.

[0003] A very common application of revenue management involves service providers accepting reservations for "date-constrained services." Date-constrained services impose transaction-specific restrictions on the days on which a buyer can use the purchased service. Examples of such restrictions include specified arrival and departure dates for airline reservations and specified check-in and check-out dates for hotel reservations. This time restriction makes it particularly difficult, especially in the hotel industry, to estimate demand and then determine optimized pricing that maximizes revenue / profit for date-constrained services. Summary of the Invention

[0004] overview An embodiment models the demand and pricing of hotel rooms. The embodiment receives historical data for multiple previous guests and generates a multinomial logit ("MNL") model with demand shock variables, which are represented using MNL utility parameters. The embodiment estimates the MNL utility parameters using likelihood maximization and determines the demand shock parameters using the estimated MNL utility parameters. The embodiment then forecasts future demand for hotel rooms based on the demand shock parameters. [Brief explanation of the drawings]

[0005] [Figure 1] FIG. 1 is a graph illustrating the endogeneity of hotel room prices. [Figure 2] FIG. 2 is a block diagram of a computer server / system according to an embodiment of the present invention. [Figure 3] FIG. 3 is a flow diagram illustrating the functionality of the room demand model module of FIG. 2 according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0006] Detailed Description Embodiments include a novel demand modeling system adapted to forecast demand for different tiers of hotel rooms based on room price and characteristics. The forecasts can be calibrated with available information about expected demand on specific days. Embodiments use this demand modeling to determine optimal pricing for hotel rooms. To consistently estimate the underlying demand model in the presence of price endogeneity, embodiments implement a multinomial logit model with time fixed effects. Embodiments correct for possible time shocks that may be driving price and demand in the same direction, and provide an intuitive way to calibrate demand forecasts when information about such future shocks is available.

[0007] Embodiments correct for price endogeneity, a widely recognized problem in the context of hotel demand estimation. Failure to correct for endogeneity leads to biased estimates. Price endogeneity arises when explanatory variables in a model are correlated with several unobserved factors that may affect the outcome variable of interest. The most prominent example of price endogeneity in the context of the hotel industry is endogenous room prices. This is because hoteliers' common practice is to set high prices during periods when high demand is expected (e.g., weekends, holidays, vacations) and low prices when demand is expected to be relatively low. Estimating price sensitivity without considering endogeneity can result in significantly underestimated price sensitivity coefficients and even have the wrong (positive) sign, because the model would conclude that, on average, high demand corresponds to periods of high prices and vice versa.

[0008] Figure 1 is a graph illustrating price endogeneity for hotel rooms. In Figure 1, the x-axis represents the average nightly room price, increasing from right to left. The y-axis represents the hotel's occupancy rate, where 1.0 corresponds to 100% of rooms being booked. For most goods and services, demand decreases as price increases. However, for many hotels, such as the example in Figure 1, price endogeneity actually reflects increasing demand as price increases, as indicated by the upward-sloping line 100 (rather than the expected downward slope).

[0009] Known solutions in industries such as hotels, airlines, and retail use an instrumental variable ("IV") approach to correct for endogeneity. However, the IV approach requires finding a valid instrumental variable, i.e., a variable that is correlated with the endogenous variable and uncorrelated with all others. To estimate the demand for a particular product, the instrumental variables may be selected as the costs of the ingredients used in the production of that product. For example, in an analysis of the demand for laundry detergent, the instrumental variables may be the price indexes for alkali and chlorine ingredients and the price of plastic used in packaging. The general idea is that changes in the prices of these ingredients will cause manufacturers to adjust the prices of final goods (e.g., laundry detergent). However, changes in the demand for final goods are not correlated with changes in ingredient prices.

[0010] If the demand system is modeled so that price affects the outcome variable linearly, the IV approach can be used quite simply; two-stage least squares ("2SLS") regression is a common approach. Other approaches use more sophisticated discrete choice models, allowing the demand model to capture the interdependencies between demand for different products offered by a company. For example, in the case of a hotel, different room types represent different options that customers can choose from when booking, and the demand model may capture the demand elasticity of a particular room type to changes in the price of other room types. In discrete choice models, where price affects demand nonlinearly, the 2SLS procedure is not applicable; approaches such as "BLP" and control functions may be used, depending on the model formulation.

[0011] However, because IV solutions cannot be implemented without valid instrumental variables, their practical application is limited by the difficulty of finding suitable instrumental variables, especially in the hospitality setting. Generally, all known solutions that correct for endogeneity require the use of instrumental variables. For some known solutions that do not use IV, the range of data required for estimation typically includes detailed price and competitor demand data, which are costly or even impossible to collect.

[0012] In contrast to known solutions for modeling demand for hotel rooms, embodiments add time shocks directly to the model, allowing for these time shocks to simultaneously affect price and demand. Embodiments generate consistent estimates of the model parameters. The shocks themselves are not observable and are estimated together with the remaining model parameters. The number of time periods typically spanned by a data set can be very large, and estimating the model directly would result in a very large number of parameters to estimate. Therefore, embodiments transform the original problem so that only a few parameters are estimated simultaneously, with all time shock parameters being estimated in a separate subsequent step. Therefore, the computational power required to estimate the model is significantly reduced, making application of the model very practical and improving computer performance.

[0013] 2 is a block diagram of a computer server / system 10 according to an embodiment of the present invention. While shown as a single system, the functionality of system 10 may be implemented as a distributed system. Additionally, the functionality disclosed herein may be implemented on separate servers or devices that may be coupled to each other via a network. Furthermore, one or more components of system 10 may not be included. For example, when implemented as a web server or cloud-based functionality, system 10 may be implemented as one or more servers, and no user interface, such as a display, mouse, or the like, is required.

[0014] System 10 includes a bus 12 or other communication mechanism for communicating information and a processor 22 coupled to bus 12 for processing information. Processor 22 may be any type of general-purpose or special-purpose processor. System 10 further includes memory 14 for storing information and instructions executed by processor 22. Memory 14 may be comprised of any combination of random access memory (“RAM”), read-only memory (“ROM”), static storage such as a magnetic or optical disk, or other types of computer-readable media. System 10 further includes a communication device 20, such as a network interface card, to provide access to a network. Thus, a user may interface with system 10 directly, remotely through a network, or otherwise.

[0015] Computer-readable media can be any available media that can be accessed by processor 22, including both volatile and nonvolatile media, removable and non-removable media, and communication media. Communication media may include computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.

[0016] Processor 22 is further coupled to a display 24, such as a liquid crystal display ("LCD"), via bus 12. A keyboard 26 and a cursor control device 28, such as a computer mouse, are further coupled to bus 12 to allow a user to interface with system 10.

[0017] In one embodiment, memory 14 stores software modules that provide functionality when executed by processor 22. These modules include an operating system 15 that provides operating system functionality for system 10. The modules further include a room demand model module 16 that forecasts hotel room demand and generates a room demand model for maximizing hotel room revenue, as well as all other functionality disclosed herein. System 10 may be part of a larger system. Thus, system 10 may include one or more additional function modules 18 to include additional functionality, such as functionality of a property management system (“PMS”) (e.g., “Oracle Hospitality OPERA Property” or “Oracle Hospitality OPERA Cloud Services”) or an enterprise resource planning (“ERP”) system. A database 17 is coupled to bus 12 to provide centralized storage for modules 16 and 18, storing guest data, hotel data, transaction data, and the like. In one embodiment, database 17 is a relational database management system (“RDBMS”) capable of managing stored data using structured query language (“SQL”). In one embodiment, a specialized point of sale ("POS") terminal 99 generates transactional and historical sales data (e.g., data regarding hotel guest / customer transactions) that is used to perform the optimization. The POS terminal 99 itself may include additional processing functionality for performing room allocation optimization according to one embodiment and may operate alone or in conjunction with the other components of Figure 2 as a specialized hotel room price optimization system.

[0018] In one embodiment, especially when there are a large number of hotel locations, a large number of guests, and a large amount of historical data, database 17 is implemented as an in-memory database ("IMDB"). An IMDB is a database management system that relies primarily on main memory for computer data storage. This contrasts with database management systems that utilize disk storage mechanisms. Main-memory databases are faster than disk-optimized databases because disk access is slower than memory access, and the internal optimization algorithms are simpler and execute fewer CPU instructions. Accessing data in memory provides faster and more predictable performance than disk, since seek times are eliminated when querying data.

[0019] In one embodiment, when database 17 is implemented as an IMDB, it is implemented based on a distributed data grid. A distributed data grid is a system in which a collection of computer servers cooperate as one or more clusters to manage information and related operations, such as computation, in a distributed or clustered environment. A distributed data grid can be used to manage application objects and data shared across servers. Distributed data grids provide low response time, high throughput, predictable scalability, continuous availability, and information reliability. In a particular example, a distributed data grid, such as Oracle's "Oracle Coherence" data grid, stores information in-memory to achieve higher performance and uses redundancy in keeping copies of that information synchronized across multiple servers to ensure system resilience and continuous availability of data in the event of a server failure.

[0020] In one embodiment, system 10 is a computing / data processing system that includes an application or collection of distributed applications for an enterprise organization and may also perform logistics, manufacturing, and inventory management functions. Application and computing system 10 may be configured to operate with or implemented as a cloud-based networking system, a software-as-a-service ("SaaS") architecture, or other type of computing solution.

[0021] As mentioned above, accurately estimating the underlying demand model for hotel rooms is an essential component of product recommendations and personalized pricing. Biased estimated demand model coefficients can lead to suboptimal pricing of hotel rooms and services, resulting in revenue losses for hotel operators.

[0022] Baseline MNL demand model

[0023]

number

[0024]

number

[0025] If the underlying model is accurately specified, the model parameters can be consistently estimated by maximizing the likelihood function. However, problems arise when the model has some price endogeneity. In the hotel context, a typical practice for hoteliers is to raise prices when there are several positive demand shocks and lower prices when demand is low in order to attract customers. Estimating a standard MNL model for data with such endogeneity can result in price sensitivity coefficients that are underestimated and may even have the wrong sign. Furthermore, even if prices are completely exogenous, estimates can still be biased if the model has several common shocks that simultaneously affect the utility of customer groups, although this is usually less of a concern than price endogeneity.

[0026] Therefore, embodiments implement the following modifications of the traditional MNL model to address possible endogeneity / common shock issues. In the following, we use "room" instead of the more generic "option" to assume a hotel-specific embodiment. Other terminology may be used in other embodiments applicable to airplane seats, rental cars, or other date-constrained services.

[0027]

number

[0028] Thus, consumers planning a trip on a particular date experience a demand shock that induces proportional changes in demand for all rooms. The ratio of demand for any two options must remain the same if price remains constant over time. Specifically, if the normal demand for first-class rooms is 100 units and the normal demand for suites is 4 units, this assumption implies that if demand for first-class rooms increases to 150 units, the demand for suites must increase to 6 if price remains constant. Therefore, when determining the price and utility of such rooms and considering price changes, the demand for different units will follow a proportional relationship.

[0029] Likelihood maximization

[0030]

number

[0031]

number

[0032] Although embodiments have M+T+1 parameters in the model to maximize the log-likelihood function, this is generally not a scalable operation, especially as T becomes large. However, embodiments can transform the optimization into solving a system of M (nonlinear) equations.

[0033]

number

[0034]

number

[0035]

number

[0036]

number

[0037]

number

[0038] Estimation Algorithm

[0039]

number

[0040] Berkson-Theil based estimation In another embodiment, instead of solving the system of equations disclosed above using likelihood maximization, we estimate a baseline model with time shocks (aggregate data, rather than individual choices) with ordinary least squares minimization. This embodiment is based on the well-known "Berkson-Theil" method, generalized and modified to include time-specific demand shocks.

[0041] The estimation procedure in this embodiment is only asymptotically equivalent to likelihood maximization, and should therefore only be used when the number of observations within each period is sufficiently large. Furthermore, this estimation procedure can only be used when non-zero purchases are observed for all categories in all periods, whereas likelihood maximization can handle cases where purchases are zero for some product categories in some periods. Finally, as a baseline algorithm, this approach requires aggregate data.

[0042]

number

[0043]

number

[0044] Individual Selection

[0045]

number

[0046]

number

[0047]

number

[0048] Clustering

[0049]

number

[0050]

number

[0051]

number

[0052] regularization

[0053]

number

[0054] Figure 3 is a flow diagram illustrating the functionality of the room demand model module 16 of Figure 2 according to an embodiment. In one embodiment, the functionality of the flow diagram of Figure 3 is implemented by software stored in memory or other computer-readable or tangible medium and executed by a processor. In other embodiments, the functionality may be performed by hardware (e.g., through the use of application specific integrated circuits ("ASICs"), programmable gate arrays ("PGAs"), field programmable gate arrays ("FPGAs"), etc.), or any combination of hardware and software.

[0055] At 302, historical reservation data and guest information is received from an input dataset / database 17. In one embodiment, input dataset 17 is Oracle's "OPERA" database, which contains details about guests and available rooms for a single hotel or a group of hotels, such as a hotel chain. In other embodiments, a database of guest and room data for any type of PMS can be used. In an embodiment, input dataset 17 is received by electronic communication from a computing device under the control of a hotel operator and then parsed by system 10 to extract information necessary for subsequent functions disclosed below.

[0056] At 304, a multinomial logit model with demand shock variables is constructed. In one embodiment, this model is disclosed above in equation (5).

[0057] At 306, we express the demand shock variables through other parameters of the MNL utility function in equation (7).

[0058] At 308, the MNL utility parameters are estimated using likelihood maximization with a limited number of variables as specified in equations (9) and (10).

[0059] At 310, the demand shock parameters are calculated from the estimated MNL utility parameters according to equation (7).

[0060] At 312, future demand periods with potentially high demand variability are identified.

[0061]

number

[0062]

number

[0063] In response to the pricing, embodiments accept reservations based on the optimized pricing and facilitate hotel stays based on the reservations. The optimized pricing may be stored in a database in the form of specialized data. Facilitating the hotel stay may include transmitting the specialized data to other specialized devices that use the specialized data, such as automatically operating a device using the specialized data to encode a hotel key or automatically programming a hotel room door lock using the specialized data.

[0064] In addition to the functionality of FIG. 3 , embodiments use the determined optimal pricing to store and update a database that provides prices for online services. These updates can occur frequently (e.g., multiple times per day or hourly), automatically modifying electronic devices based on the revised prices. Furthermore, in embodiments, greater hotel utilization may result in additional services being utilized at the hotel. Furthermore, in embodiments, the optimized prices are transmitted over a network, causing other computing devices / servers to modify the prices in the pricing database according to the revised, optimized prices.

[0065] In a slightly different paradigm / embodiment, commonly known as explore-exploit, prices offered to a subset of customers are randomly deviated from the optimal price (exploration phase) to improve demand learning. This allows the model to be trained on a greater variety of responses, resulting in more robust coefficient estimates. A new optimal price is then calculated and offered through the reservation system, and the loop is repeated. Each new customer response to this offer is stored in a database and used to continually retrain the model.

[0066] Because a typical hotel chain receives reservations through multiple channels, such as global distribution systems (e.g., Amadeus), online travel agencies (e.g., Expedia, Booking.com), and business travel management companies (e.g., CWT (formerly Carlson Wagonly Travel)), as well as through hotel chain-operated channels such as voice, web, and email reservations, embodiments of the present invention are implemented to interface with various application programming interfaces ("APIs") to make the collection of reservation data as independent as possible from the booking user's hardware. However, in some embodiments, to expedite the delivery of personalized offers and recommendations provided by the present invention to hotel operators, embodiments are implemented to run on multiple hardware devices, from handheld mobile phones and computer pads to desktops used by booking agents.

[0067] Comparison with the MNL model without time shocks Below is a simulation comparison between a demand model based on a simple MNL model (referred to as "MNL") and an embodiment of the present invention (referred to as "TS") that incorporates time shocks into MNL. Several price / choice datasets are simulated to illustrate the advantages of the embodiment and show how robust the embodiment is to various misspecifications of the model.

[0068]

number

[0069]

number

[0070]

number

[0071] To eliminate the possibility that randomness in the samples could cause bias in the estimation, all estimates were averaged over 100 different samples, the results of which are shown in Table 1 below.

[0072]

number

[0073]

number

[0074] Furthermore, while various price endogeneities do not bias the estimates when the underlying model is accurate (specifications 2a and 2b), they clearly bias the MNL significantly.

[0075] Furthermore, if there is endogeneity in prices and the model is misspecified (specifications 3a and 3b), anything can happen. From the estimations, TS appears to perform a little better than MNL, but this really depends on the magnitude of the endogeneity and how misspecified the model is.

[0076]

number

[0077] Furthermore, the embodiment is tested on a real hotel demand dataset. Compared to the results of a simple MNL model that does not correct for price endogeneity, the magnitude of the estimated price sensitivity coefficients in the embodiment is up to 30% higher.

[0078] As disclosed, embodiments accurately determined demand for hotel rooms (or other date-constrained services) by (1) generating a novel model that accounts for price endogeneity without requiring instrumental variables, (2) providing an easy transformation of the original problem that significantly reduces the number of parameters that must be simultaneously estimated and requires little computational power, and (3) not requiring knowledge of the number of non-purchasing customers or even additional assumptions about market size for consistent estimation of price sensitivity coefficients (and the relative value of room categories). Consistent estimation of the underlying demand model is crucial for price optimization, product recommendations, upsell recommendations, etc.

[0079] Although several embodiments have been specifically shown and / or described herein, it will be understood that modifications and variations of the disclosed embodiments are encompassed within the scope of the appended claims in accordance with the above teachings without departing from the spirit and intended scope of the invention.

Claims

1. 1. A method for modeling demand and pricing for hotel rooms, said method comprising: receiving, by one or more processors, historical data relating to a plurality of previous guests; the one or more processors generating a multinomial logit (MNL) model with demand shock variables; [Equation 1] The one or more processors estimate the MNL utility parameters by maximizing a likelihood function, wherein the likelihood function is: [Equation 2] where y it is the number of customers arriving at time t, and maximizing the likelihood function is k and taking a first derivative of the likelihood function with respect to β and a second derivative of the likelihood function with respect to β, and the MNL utility parameters are k (k=1,...,M) and β, and the method further comprises: The one or more processors may further estimate the MNL utility parameters by t determining the demand shock variable by substituting into the third derivative of the likelihood function with respect to and the one or more processors forecasting, for a future demand period, the future demand for the hotel rooms based on the MNL model defined by the demand shock variables determined by using certain data from the historical data corresponding to a past demand period similar to the future demand period.

2. The one or more processors calculate the future demand and the price p for the kth option. k Calculating hotel revenue based on The method of claim 1 , further comprising the one or more processors finding an optimal set of prices for the hotel rooms that optimizes the hotel revenue.

3. The method of claim 1 or 2, wherein the likelihood function comprises an aggregate likelihood function defined using aggregate data corresponding to selections by multiple customers.

4. The method of claim 1 or 2, wherein the likelihood function comprises a likelihood function of an individual selection.

5. 5. The method of claim 1, wherein the historical data comprises customer characteristics, the method further comprising the one or more processors clustering the customer characteristics, and the likelihood function is defined for each cluster.

6. the one or more processors offering a first hotel room at an optimized price based on the optimal price set; the one or more processors accepting a selection of the first hotel room; In response to the selection, the one or more processors reserve the first hotel room for the first customer at the optimized price; 3. The method of claim 2, further comprising: in response to the reservation, the one or more processors encoding a hotel room key corresponding to the first hotel room by using a hotel room key machine. [Request Item 7] [Number 3]

8. A program comprising instructions which, when executed by one or more processors, cause the processors to perform the method of any one of claims 1 to 7.

9. 1. A hotel room pricing system comprising: one or more processors coupled to the stored instructions; a database storing historical data relating to a plurality of previous guests; The processor is configured to model demand for hotel rooms, and modeling the demand for hotel rooms comprises: generating a multinomial logit (MNL) model with demand shock variables; [Equation 4] estimating the MNL utility parameters by maximizing a likelihood function, the likelihood function being: [Equation 5] where y it is the number of customers arriving at time t, and maximizing the likelihood function is k and taking a first derivative of the likelihood function with respect to β and a second derivative of the likelihood function with respect to β, and the MNL utility parameters are k (k=1,...,M) and β, and modeling the demand for hotel rooms further comprises: The estimated MNL utility parameters are denoted by γ t determining the demand shock variable by substituting into the third derivative of the likelihood function with respect to and forecasting, for a future demand period, future demand for the hotel rooms based on the MNL model defined by the demand shock variables determined by using certain data from the historical data corresponding to a past demand period similar to the future demand period.

10. Modeling the demand for hotel rooms comprises: The future demand and the price p for the kth alternative k Calculating hotel revenue based on 10. The system of claim 9, further comprising the one or more processors finding an optimal set of prices for the hotel rooms that optimizes the hotel revenue.

11. 11. The system of claim 9 or 10, wherein the likelihood function comprises an aggregate likelihood function defined using aggregate data corresponding to selections by a plurality of customers.

12. 11. The system of claim 9 or 10, wherein the likelihood function comprises a likelihood function of an individual selection.

13. It also has a hotel room key machine. The processor: offering a first hotel room at an optimized price based on the optimal price set; Accepting a selection of the first hotel room; In response to the selection, reserving the first hotel room for the first customer at the optimized price; 11. The system of claim 10, further configured to encode a hotel room key corresponding to the first hotel room by using the hotel room key machine in response to the reservation. [Request Item 14] [Number 6]

Images