This is the second in a series of blog posts on deposit price elasticity, focusing on the modelling data requirements.
There are several different modelling techniques and approaches to measure deposit price elasticity, which is dependent on the actual business problem and model usage. The exact data requirements might need to be amended to account for the modelling technique, but a large number of data items are consistent across all approaches.
As with all modelling projects, it is good to initially take a step back and think about what type of information you would expect to be predictive, impact deposit price elasticity, and make you move your savings across different products and / or financial institutions.
We would suggest that considering the below fields would be best practice for a deposit price elasticity model development:
- Product details: Historical interest rate of the product(s) to be modelled, this would include retention or bonus period offers, which might impact demand.
- Competitor pricing: To understand what products and interest rates were available on the market, so these can be compared against the product(s) that are to be modelled.
- Existing customer product holdings: What are the volume and value of accounts reaching product maturity, or end of a bonus period, which would impact future demand? Consider how customer segments previously behaved at the end of the maturity period.
- Seasonality trends: These can be obtained from historical demand, or market trends, which might be impacted by the end of the tax year, or seasonal events.
- Marketing data: Is the product advertised on aggregator websites? What historical marketing actions have taken place that might impact demand, and bias historical data?
- Macro-economic data: Changes to the central bank lending rate, or trends in the household savings ratio.
- Predictive space: How much product pricing has varied relative to the market, over the modelling period, and how to best control for this, where the predictive space is limited.
- Competitor set: How to compare products against the market — either the full market, and / or a subset of “like for like” competitors.
- Weighted Average Market Price (WAMP): Rather than considering the best / average market price, alternative metrics such as WAMP can be generated. Different weighting factors can also be explored. For example, share of deposits, customers, or branches within a particular market are all candidates for WAMP.
- Pricing relative to competitors: What type of characteristics need to be generated, to compare product pricing to the market.
- Pricing relative to neighbouring products: What are the neighbouring products that might be at risk of product cannibalisation? What type of characteristics need to be generated to measure this?
It is also important to consider exceptional events that might have taken place, which might impact demand. These can be used to help understand outliers in the data, exceptional events that might be system issues, or regulation changes.
As with all modelling problems, additional data items can be considered that might provide additional insights. The above list highlights what we would start with, but if any of the above data items are not available, models can still be developed on a subset of the above data fields. Not having all this information available might mean that assumptions might need to be incorporated within version 1 of the models, and data capture processes improved, so the data items can be considered in future iterations.
In my next post, I will cover the different modelling techniques, approaches, and challenges. I invite you to find out more about FICO’s deposit price elasticity modelling and optimisation.