A simulation based analysis of price elasticity of demand
| Název práce v češtině: | Analýza cenové elasticity poptávky založená na simulacích |
|---|---|
| Název v anglickém jazyce: | A simulation based analysis of price elasticity of demand |
| Klíčová slova: | price elasticity of demand, artificial neural network, agent-based model |
| Klíčová slova anglicky: | cenová elasticita poptávky, umělé neuronové sítě, agentové modely |
| Akademický rok vypsání: | 2018/2019 |
| Typ práce: | diplomová práce |
| Jazyk práce: | angličtina |
| Ústav: | Institut ekonomických studií (23-IES) |
| Vedoucí / školitel: | doc. PhDr. RNDr. Josef Stráský, Ph.D. |
| Řešitel: | skrytý - zadáno vedoucím/školitelem |
| Datum přihlášení: | 19.05.2019 |
| Datum zadání: | 17.07.2019 |
| Datum a čas obhajoby: | 16.06.2020 09:00 |
| Datum odevzdání elektronické podoby: | 07.05.2020 |
| Datum proběhlé obhajoby: | 16.06.2020 |
| Oponenti: | RNDr. Michal Červinka, Ph.D. |
| Kontrola URKUND: | |
| Zásady pro vypracování |
| The main goal of this thesis is to explore the field of modelling the movement of the customers through the FMCG (fast moving consumer goods) website. The e-commerce retailers are able to collect much more comprehensive information, with the inclusion of an additional level of detail in the shopping data, extending the traditional transaction level by the timestamp of the moment when every single product was added to the cart. With the use of such dataset and the website map, we aim to develop a probabilistic model that would be able to describe the website flow of customers. Furthermore, we intend to extend such methodology by including external factors (such as promotional activity, or secondary placements), thus modelling the conditional probabilities of shopping sequences. Finally, we plan to develop a simulation that would be able to visualize the customer paths under different events, based on conditional probabilities. |
| Seznam odborné literatury |
| Artemenko, V. (2013): “Web analytics in e-learning: Agent-based and neural
network approaches.” Proceedings of the 2013 IEEE 7th International Conference on Intelligent Data Acquisition and Advanced Computing Systems, IDAACS 2013 2(September): pp. 774–780. Kwark, Y., G. M. Lee, P. A. Pavlou, & L. Qiu (2016): “The spillover effects of user-generated online product reviews on purchases: Evidence from clickstream data.” 2016 International Conference on Information Systems, ICIS 2016 . Lang, T. & M. Rettenmeier (2017): “Understanding Consumer Behavior with Recurrent Neural Networks.” MLRec 2017 (Approach 1): pp. 1–8. Li, J., S. Netessine, & S. Koulayev (2018): “Price to compete... with many: How to identify price competition in high-dimensional space.” Management Science 64(9): pp. 4118–4136. Montgomery, A. L., S. Li, K. Srinivasan, & J. C. Liechty (2004): “Mod- eling online browsing and path analysis using clickstream data.” Marketing Science 23(4): pp. 579–595. Wang, S., C. Liu, X. Gao, H. Qu, & W. Xu (2017): “Session-Based Fraud Detection in Online E-Commerce Transactions Using Recurrent Neural Networks.” volume 1, pp. 241–252. Minih, V. & K. Kavukcuoglu (2017): “Methods and apparatus for reinforcement learning.” |
| Předběžná náplň práce |
| Motivation
The study of price elasticity of demand (PED) is one of the fundamental pieces of microeconomic theory as it aspires to characterize customer behaviour and how it shifts with changing prices. This concept is an unavoidable basis of any price-sensitive demand function, having its part in the vast range of economic models. The behaviour described by the elasticity can have a profound effect on the taxation, wage legislative and any other price affecting policies. Taking the taxation as an example, the introduction of the same tax for an elastic and inelastic product would lead to wildly different results. For a product with high elasticity, a large reaction of demand to a change in price, the demand is likely to sharply decline, resulting in a significant shape of customer behaviour. On the other hand, the same policy imposed on a product with low elasticity would result in a much lower reaction of the demand, resulting in significant tax income. Therefore, the failure to consider this relationship can result in unpredictable results of the new policies. The price elasticity is not only one of the building blocks of economic research, as this notion is highly relevant to the business. One of the key areas of any company is the development of a pricing strategy, a process of setting the price of its products on the market, and the price elasticity is one of the obvious choices of the method to estimate the effects of price alterations on the revenues (or profit). Understanding these practices is crucial not only for the companies and their competition but as well for the decision-makers developing the policies to regulate pricing activities of those companies (such as price-cap regulations). Hypotheses The main goal of this thesis is to explore the field of modelling the price elasticity of demand using the sequences of purchasing patterns, an ordered log of products purchased. We aim to develop a probabilistic model of this behaviour that is consequently used in agent-based simulations. Our hypotheses revolve around the validation of the developed methodology, notably: 1. The method can estimate the elasticity coefficients at any individual price point. 2. The method takes into account the price position of other products (cross-elasticity). 3. The method can be flexibly extended with various external variables. Methodology The focus of our work will be on developing a novel estimation method based on simulations from agent-based models (ABM), focusing on relaxing the assumption of global price elasticity. We build the ABM simulation using a model-based approach, instead of initializing the probability distributions of various actions the agent can take. In the first step, we prepare a model of the user behaviour, that predicts the next purchase of the customer given an external situation (price strategy and products already bought, in our case). This model is then used in the agent-based model, assembling the simulated transactions. Utilising the resulting replicas, we are able to analyze the elasticity and compare it across diverse setups. Expected Contribution The current approaches mostly utilize linear models, providing a fixed (global) elasticity estimate. As a result, the studies are based on a hidden premise of constant elasticity (even in the extreme values), which is a significant assumption. And while there are many extensions to these methods, such as vector autoregression models (VAR) or approaches used in panel data analysis, they are able to consider only a limited portion of the external situations that can affect the elasticity. In the real-world setup, the price elasticity can differ in every distinct price point. It depends not only on the own price but on the prices of various diverse products and promotional activity and the metric is likely to differ for buyers from various regions, or different store types (e.g. online or physical). We focus on developing a completely new methodology for analysis of PED. With the proposed approach, we aim to increase the accuracy of the estimation of the coefficients of price elasticity, notably allowing for analysis of local coefficients (the elasticity in each price point), and by presenting a robust framework allowing for the inclusion of additional predictive variables. 1. Introduction 2. Literature overview 3. Aims 4. Data description and preparation 5. Methodology 6. Result 7. Conclusion |
| Předběžná náplň práce v anglickém jazyce |
| Motivation
The study of price elasticity of demand (PED) is one of the fundamental pieces of microeconomic theory as it aspires to characterize customer behaviour and how it shifts with changing prices. This concept is an unavoidable basis of any price-sensitive demand function, having its part in the vast range of economic models. The behaviour described by the elasticity can have a profound effect on the taxation, wage legislative and any other price affecting policies. Taking the taxation as an example, the introduction of the same tax for an elastic and inelastic product would lead to wildly different results. For a product with high elasticity, a large reaction of demand to a change in price, the demand is likely to sharply decline, resulting in a significant shape of customer behaviour. On the other hand, the same policy imposed on a product with low elasticity would result in a much lower reaction of the demand, resulting in significant tax income. Therefore, the failure to consider this relationship can result in unpredictable results of the new policies. The price elasticity is not only one of the building blocks of economic research, as this notion is highly relevant to the business. One of the key areas of any company is the development of a pricing strategy, a process of setting the price of its products on the market, and the price elasticity is one of the obvious choices of the method to estimate the effects of price alterations on the revenues (or profit). Understanding these practices is crucial not only for the companies and their competition but as well for the decision-makers developing the policies to regulate pricing activities of those companies (such as price-cap regulations). Hypotheses The main goal of this thesis is to explore the field of modelling the price elasticity of demand using the sequences of purchasing patterns, an ordered log of products purchased. We aim to develop a probabilistic model of this behaviour that is consequently used in agent-based simulations. Our hypotheses revolve around the validation of the developed methodology, notably: 1. The method can estimate the elasticity coefficients at any individual price point. 2. The method takes into account the price position of other products (cross-elasticity). 3. The method can be flexibly extended with various external variables. Methodology The focus of our work will be on developing a novel estimation method based on simulations from agent-based models (ABM), focusing on relaxing the assumption of global price elasticity. We build the ABM simulation using a model-based approach, instead of initializing the probability distributions of various actions the agent can take. In the first step, we prepare a model of the user behaviour, that predicts the next purchase of the customer given an external situation (price strategy and products already bought, in our case). This model is then used in the agent-based model, assembling the simulated transactions. Utilising the resulting replicas, we are able to analyze the elasticity and compare it across diverse setups. Expected Contribution The current approaches mostly utilize linear models, providing a fixed (global) elasticity estimate. As a result, the studies are based on a hidden premise of constant elasticity (even in the extreme values), which is a significant assumption. And while there are many extensions to these methods, such as vector autoregression models (VAR) or approaches used in panel data analysis, they are able to consider only a limited portion of the external situations that can affect the elasticity. In the real-world setup, the price elasticity can differ in every distinct price point. It depends not only on the own price but on the prices of various diverse products and promotional activity and the metric is likely to differ for buyers from various regions, or different store types (e.g. online or physical). We focus on developing a completely new methodology for analysis of PED. With the proposed approach, we aim to increase the accuracy of the estimation of the coefficients of price elasticity, notably allowing for analysis of local coefficients (the elasticity in each price point), and by presenting a robust framework allowing for the inclusion of additional predictive variables. 1. Introduction 2. Literature overview 3. Aims 4. Data description and preparation 5. Methodology 6. Result 7. Conclusion |