Dynamic pricing patterns on an Internet distribution channel: the case study of Bilbao's hotels in 2013
Download as PPTX, PDF1 like119 views
The document presents a study conducted to analyze dynamic pricing patterns of hotels in Bilbao in 2013, utilizing a data collection methodology involving a custom data scraping bot. It identifies key findings regarding price changes, their frequency, magnitude, and the respective duration, as well as distinguishing between vertical and diagonal pricing patterns. The research concludes with a proposal for future work to explore the causes behind these pricing patterns and develop predictive models.
![ENTER 2015 Research Track Slide Number 9
Methodology
3.3 Temporal relationships in the data
1. [target date]. Prices offered by the same hotel corresponding to the same target date
all refer to the same product. However, prices corresponding to the same target date
but offered by different hotels are prices for competing products.
2. [collection date]. At any one instant in time, only prices with the same collection
date are simultaneously available, both for the clients to book a room and for the
hoteliers to increase or decrease the price.
3. The difference in days between the two dates.
The analysis of the data has been carried out with these temporal relationships as a key
factor.](https://image.slidesharecdn.com/enter15dynamicpatternsv3kepa-150403054820-conversion-gate01/85/Dynamic-pricing-patterns-on-an-Internet-distribution-channel-the-case-study-of-Bilbao-s-hotels-in-2013-9-320.jpg)
Dynamic pricing patterns on an Internet distribution channel: the case study of Bilbao's hotels in 2013
- 1. ENTER 2015 Research Track Slide Number 1 Dynamic pricing patterns on an Internet distribution channel: the case study of Bilbao's hotels in 2013 Noelia Oses, Jon Kepa Gerrikagoitia and Aurkene Alzua-Sorzabal CICtourGUNE Donostia - San Sebastián, Spain http://www.tourgune.org
- 2. ENTER 2015 Research Track Slide Number 2 1. Introduction 2. Related Work 3. Methodology 4. Results 5. Conclusion and future work
- 3. ENTER 2015 Research Track Slide Number 3 Introduction • The Internet has become part of society’s basic infrastructure. More and more activities are taking place through the Internet, leaving behind digital footprints, which can be detected and measured in real time. This also applies for tourism accommodations and their characteristics. This information can be collected from the internet by software programs named crawlers. (Heerschap et al, 2014). • The aim of contributing to new technologies and methods of tourism measurement has been included at CICtourGUNE´s research agenda. A technological platform and a robust methodology to get online hotel price and availability named Dynamic Pricing Monitor has been implemented (Gerrikagoitia, Alzua et al, 2011)
- 4. ENTER 2015 Research Track Slide Number 4 Introduction • The motivation of the research is to understand and determine the current dynamic pricing practices of the hotels in Bilbao. • A data analysis based on CICtourGUNE´s accommodation industry data assets has been conducted with hotels in Bilbao during 2013. • The main contributions of the research work have been: • Definition of monitoring, measurement and visualization framework at micro level (property) • The identification of different dynamic princing patterns • The working hypothesis relating patterns and causality
- 5. ENTER 2015 Research Track Slide Number 5 Related Work • Kimes & Wirtz (2003) define Revenue Management as the application of information systems and pricing strategies to allocate the right capacity to the right client at the right time to maximise revenue. Characteristics: perishable inventory, limited capacity, demand volatility, microsegmentation, advance booking availability, and high fixed costs (Ivanov & Zhechev, 2012). • Dynamic pricing is a pricing policy based on time, where prices change depending on the day on which the reservation is processed. Abrate et al. (2012) reached the conclusion that more than 90% of the prices change with time depending, mostly, on the type of client (business or leisure) and the star rating. • Hoteliers are using different strategies to improve their revenue. A comparative analysis revealed significant differences in average prices due to the segmentation of the market as a function of the commercialisation channel (Tso & Law, 2005). • Customers make decision to purchase flight tickets and hotel rooms based on various factors such as the quality of the information (Wong, 2005), the time/date, past experiences (Kim & Kim, 2004), and the frequency (Magnini & Karande, 2011). But the factor that influences the hotel choice most is the price (Lockyer, 2005; Tanford et al., 2012; Tso & Law, 2005).
- 6. ENTER 2015 Research Track Slide Number 6 Related Work It is important for hoteliers to know the level of demand and the prices offered by competitors to define dynamic pricing policies. • Official Statistics regularly calculate hotel pricing indices aggregated by time and geography. • Marketing Decision Support Systems (MDSS) were developed based in IT providing information to react to competitors’ and market changes quickly. • In 1998, Wöber (1998) presented TourMIS, an online MDSS tool for tourism and accommodation management. • In 2009, Walchhofer et al. (2009b) introduced SEMAMO, a market surveillance system combining semi-supervised techniques with semantic models. • In 2011, CICtourGUNE created the Dynamic Pricing Monitor (Gerrikagoitia, Alzua et al, 2011) as a technology and science infraestructure with contributions to industry (Alzua et al, 2013) (Ibarguren et al, 2014), official statistics and public bodies (Roman et al, 2013). • Currently, BeonPrice has released a system called ’Analyzer’ which allows a company to monitor competitors’ price changes in real time Additionally to the historic and real-time information, making predictions is an essential tool in the tourism domain (Song and Li, 2008) (Haensel and Koole, 2011)
- 7. ENTER 2015 Research Track Slide Number 7 Methodology / Procedure 3.1 Data collection methodology • A data scrapping bot customised for booking.com has been used to collect the data • The bot collects prices offered for a single overnight stay in a ’Double or Twin room’ for Bilbao. • Hotel room prices change in real-time so information must be gathered periodically. • The sampling method followed to collect the prices is based on that of Abrate et al. (2012) collecting fares for hotel room reservations 1, 2, 4, 7, 15, 22, 30, 45, 60, and 90 days in advance of the date of the overnight stay (i.e. the target date). Extending this approach, the bot used collected prices 1-28, 45, 60, and 90 days.
- 8. ENTER 2015 Research Track Slide Number 8 Methodology / Procedure 3.2 Data • The collected variables are: hotel name, collection date (i.e. the date when the price is collected form the website), target date (i.e. the date of the overnight stay), price, and hotel’s star rating. • There are 316.391 observations for 2013, 9.151 (0,2%) of which have the price missing. There were 32 hotels in Bilbao operating on booking.com during 2013, distributed as follows : • 1-star: 3 hotels • 2-star: 7 hotels • 3-star: 6 hotels • 4-star:12 hotels • 5-star: 4 hotels • Target dates go from 01/01/2013 to 31/12/2013 • The name ’lag-day’ is used here to refer to the difference in days between the collection date and the target date.
- 9. ENTER 2015 Research Track Slide Number 9 Methodology 3.3 Temporal relationships in the data 1. [target date]. Prices offered by the same hotel corresponding to the same target date all refer to the same product. However, prices corresponding to the same target date but offered by different hotels are prices for competing products. 2. [collection date]. At any one instant in time, only prices with the same collection date are simultaneously available, both for the clients to book a room and for the hoteliers to increase or decrease the price. 3. The difference in days between the two dates. The analysis of the data has been carried out with these temporal relationships as a key factor.
- 10. ENTER 2015 Research Track Slide Number 10 3.4 Data visualisation • row: complete series of prices of a hotel for a target date • column: collection date • block: rows corresponding to the same target date but different hotels Methodology
- 11. ENTER 2015 Research Track Slide Number 11 • Price changes (increase or decrease) are visualised with respect to the previous price in the series • The three temporal relationships correspond to rows, columns and step-wise diagonals in the visualisation. 3.4. Data visualisation Methodology
- 12. ENTER 2015 Research Track Slide Number 12 Results The objective of the analysis is to explore the three temporal relationships in order to explain hotel price dynamics. 4.1 Number of price changes per series. How often the price is changed for a target date. 4.2 Magnitude of individual changes. How much is the increase or decrease in prices? When does it happen? 4.3 Aggregate change in price series. Is the final price increased or decreased after the changes? 4.4 Duration of price changes within a series. How long remains a price change? When does it happen? 4.5 Relative location of price changes. How many changes occur by collection date, target date and lag day? The importance of the days changes are performed, the day the product is “consumed” and the period before the product is “lapsed”.
- 13. ENTER 2015 Research Track Slide Number 13 Results 4.1 Number of price changes per series - How often the price is changed for a target date. • 75% of the price series have 5 or less price changes and 19 is the maximum number of price changes in a single series • Mean number of changes by category: 3-star (4.3), 4-star (3.8), 5-star (2.8), 2-star (2.1), and 1-star (0.9). • 39% of the price series have more increases than decreases, 32% have an equal number of price increases and decreases, and only 29% have a greater number of price decreases. • Price series of hotels in categories 4- and 5-star usually have more price increases than decreases (about 51% in both cases).
- 14. ENTER 2015 Research Track Slide Number 14 • The magnitude of the price increases is slightly greater than that of price decreases. • The mean price increase is 13.27 and the mean price decrease is 12.09, both in Euro • Change magnitudes for 1-star hotels are greater than those for 2-, 3-, and 4-star hotels. • 5-star price change magnitudes are larger than those for all other categories • Price changes are of larger magnitude for weekends (i.e. Friday and Saturday). 4.2 Magnitude of individual changes. How much is the increase or decrease in prices? When does it happen? Results
- 15. ENTER 2015 Research Track Slide Number 15 • If the sum total is positive (’Up’), then the final price is higher than the starting price in the series. • As the magnitude of the aggregate change increases in absolute value, it is more likely to be a positive aggregate change than a negative one. • In 2013 most of the aggregate changes were positive 4.3 Aggregate change in price series. Is the final price (sum total of price changes per series) increased or decreased after the changes? Results
- 16. ENTER 2015 Research Track Slide Number 16 • Categories 3-, 4-, and 5-star follow the pattern of more changes price up. • Categories 1- and 2- star behave different. • The smallest aggregate change happen on Sundays and the largest on Wednesdays, followed by Saturdays. Results Distribution of aggregate change by category.
- 17. ENTER 2015 Research Track Slide Number 17 Results 4.4 Duration of price changes within a series. How long remains a price change? When does it happen? • This section studies the duration of price changes within a series, that is, for each price change the number of days that elapse before the next price change occurs within the same price series. • Generally, price decreases last longer than price increases. • By category, 1-star price changes last significantly longer that price changes in other categories. This is to be expected as category 1-star has very few price changes and, therefore, they must last longer. • For 2- and 5-star, price decreases last noticeably longer than price increases, especially for 2- star. • By month, price changes in December last longer than in other months and price increases last longer than decreases in August, September, and October. • By day of the week, price increases last longer than decreases on Friday and Saturday and the opposite is true for the other weekdays.
- 18. ENTER 2015 Research Track Slide Number 18 Results 4.5 Relative location of price changes. How many changes occur by collection date, target date and lag day? The activity in observed days, in the days the product is “consumed” and the period before the product is “lapsed” • The count of the number of price changes with regard to the three temporal relationships target date, collection date and lag-day. • Vertical and diagonal counts have peaks for price increases and decreases at certain collection dates and lag-days, whereas the horizontal counts are rather more uniform for all target dates. • This means that price changes are not uniformly distributed across dates, they occur more at some collection dates and lag-days. This suggests that prices are changed due to underlying causes or patterns. vertical horizontal diagonal collection date target date lag days
- 19. ENTER 2015 Research Track Slide Number 19 Results Examples of vertical and diagonal change patterns • Vertical pattern: Green and red vertical lines • Diagonal pattern: Green diagonal line
- 20. ENTER 2015 Research Track Slide Number 20 Conclusion and future work • The analysis has uncovered the dynamic pricing patterns that hotels in Bilbao follow during 2013. • It has shown that Bilbao’s dynamic pricing strategy on booking.com is to make contiguous changes vertically and diagonally, that is, to change the prices for different products on the same collection date or on the same lag-day. These have been named vertical and diagonal dynamic pricing patterns. • Vertical patterns include more price decreases than increases whereas all diagonal patterns of a significant length are price increases. • Only hotels of 3-star or greater category present long diagonal patterns, but vertical patterns can be found in all categories.
- 21. ENTER 2015 Research Track Slide Number 21 Conclusion and future work • The working hypothesis is that: • diagonal patterns correspond to regular hotel dynamic pricing policies and are, mostly, independent of other hotels’ actions • vertical patterns are the direct result of external factors (factors other than hotel policy such as an increase/decrease in the competition’s prices, demand, availability...). • Future work will focus on researching the causes of vertical patterns and developing a prediction model, researching the causes of changes that are not part of vertical or diagonal patterns, and analysing the meaning of missing prices and whether they are correlated with room availability.
Editor's Notes
- #7: It is important for hoteliers to know the level of demand and the prices offered by competitors to define dynamic pricing policies. These factors, together with the knowledge of their own availability, are the key elements to create a good pricing strategy. Prices change constantly, so this information must be precise, trustworthy, and in real time. There exist different approaches to support pricing policies. On the one hand, official statistics institutes regularly calculate hotel pricing indices. aggregated by time and geography. They are useful for understanding the behaviour of the market, but not for supporting decision making in real time. To fill this gap, Marketing Decision Support Systems (MDSS) were developed with the help of information technologies. These systems provide the necessary information to react to competitors’ price changes and market changes quickly. In 1998, Wöber (1998) presented TourMIS, an online MDSS tool for tourism and accommodation management, which has been used by more than 1000 users in three years (Wöber, 2003). In 2009, Walchhofer et al. (2009b) introduced SEMAMO, a market surveillance system. It works by combining semi-supervised techniques with semantic models (Walchhofer et al., 2009a). In the tourism domain, this system has the ability to extract competitors’ hotel room prices. Currently, the Spanish company BeOnPrice has released a system called ’Analyzer’ which allows a company to monitor competitors’ price changes in real time. Similarly, eRevMax’s RateTiger system allows monitoring and periodically reporting on own and competitors’ prices in multiple channels. Additionally to the historic and real-time information, making predictions is an essential tool. In the tourism domain, Song & Li (2008) have reported on the work done in this area. Particularly, Haensel & Koole (2011) developed an algorithm to make predictions that were updated dynamically with promising results. Some companies offer a performance prediction of commercial products. For example, Hotel Horizon reports on the quarterly demand predictions for hotels indifferent segments (PKF Hospitality Research, 2010).
- #8: is the standard product for hotels and it is often used to study the accommodation market (Abrate et al., 2012).
- #9: The price is a real number (in Euro) if the scrapping bot could find the hotel on the collection date, for the given target date, and there was a price for the double standard or twin room. However, the price can be missing if the bot could not find a price for this type of room for a hotel returned by the IDC.
- #10: When working with data, it is always helpful to have a good visualisation. In this particular case, it is desirable for the visualisation to display the temporal relationships in the data in a clear and straightforward way. It is also customary to have time increase from left to right in plots. Therefore, the data has been arranged in a matrix in which time increases from left to right on the columns, each column corresponding to one (collection) date. The visualisation will be the image of this matrix. All prices on the same column correspond to the same collection date. All prices on the same row correspond to the same hotel and the same target date, each having been collected on the collection date corresponding to the column on which they are. The rows will be arranged by target date in a non-decreasing order. Thus, the rows corresponding to the same target date but different hotels will be all positioned contiguously forming a block. Blocks corresponding to succeeding target dates will also be positioned contiguously. This arrangement results in prices for which the difference in days between the two dates is the same (collection date - target date) being displayed in a stepwise diagonal. Therefore, the three temporal relationships correspond to rows, columns, and step-wise diagonals in the visualisation. Prices will be visualised as an image of the price matrix arranged as explained above. Price changes will be visualised by indicating whether a price has changed (increased or decreased) with respect to the previous available price in the same price series (i.e. The price series, or row, of all prices corresponding to the same target date and the same hotel). Fig.1 provides an illustration of the visualisation of the data for the first week of June, 2013.
- #11: The data has been arranged in a matrix in which time increases from left to right on the columns, each column corresponding to one (collection) date. The visualisation will be the image of this matrix. All prices on the same column correspond to the same collection date. All prices on the same row correspond to the same hotel and the same target date, each having been collected on the collection date corresponding to the column on which they are. The rows will be arranged by target date in a non-decreasing order. Thus, the rows corresponding to the same target date but different hotels will be all positioned contiguously forming a block. This arrangement results in prices for which the difference in days between the two dates is the same (collection date - target date) being displayed in a stepwise diagonal. Therefore, the three temporal relationships correspond to rows, columns, and step-wise diagonals in the visualisation. Prices will be visualised as an image of the price matrix arranged as explained above. Price changes will be visualised by indicating whether a price has changed (increased or decreased) with respect to the previous available price in the same price series (i.e. The price series, or row, of all prices corresponding to the same target date and the same hotel). Fig.1 provides an illustration of the visualisation of the data for the first week of June, 2013.
- #12: The data has been arranged in a matrix in which time increases from left to right on the columns, each column corresponding to one (collection) date. The visualisation will be the image of this matrix. All prices on the same column correspond to the same collection date. All prices on the same row correspond to the same hotel and the same target date, each having been collected on the collection date corresponding to the column on which they are. The rows will be arranged by target date in a non-decreasing order. Thus, the rows corresponding to the same target date but different hotels will be all positioned contiguously forming a block. This arrangement results in prices for which the difference in days between the two dates is the same (collection date - target date) being displayed in a stepwise diagonal. Therefore, the three temporal relationships correspond to rows, columns, and step-wise diagonals in the visualisation. Prices will be visualised as an image of the price matrix arranged as explained above. Price changes will be visualised by indicating whether a price has changed (increased or decreased) with respect to the previous available price in the same price series (i.e. The price series, or row, of all prices corresponding to the same target date and the same hotel). Fig.1 provides an illustration of the visualisation of the data for the first week of June, 2013.
- #18: . Duration of price changes could help understand the dynamics of price changes For example, if, say, all changes lasted 7 days for a given hotel, this might indicate that this hotel only accesses its channel manager once a week and, therefore, these changes would be independent of other factors (such as competition's price changes). 4-star price change durations are rather even for price increases and decreases.
- #19: ’Horizontally contiguous’ refers to those prices corresponding to the same target date and hotel that were collected on consecutive collection dates. ’Vertically contiguous’ refers to those prices corresponding to the same hotel and consecutive target dates that were collected on the same collection date. ’Diagonally contiguous’ refers to those prices corresponding to the same hotel and consecutive target dates that have the same lag-day. Price changes are said to be ’contiguous’ when they are prices that are contiguous in any of the three possible directions defined here and they all represent a price change in the same direction (either a price increase or a decrease).