F Non-Maturity Products Model Analysis

Exploratory Data Analysis – At this stage, User can pick the historical data which will be used for model analysis in subsequent steps.

This UI lets you define as of date range, for which historical data needs to be picked. As Of Date (Frequency) – This helps you define data points frequency, which could be daily, monthly, bi-weekly etc. More frequent/granular data, helps you enhance your model quality. Here, you need to make sure not to select frequency which is less than frequency at which historical data is available. So, if you have loaded monthly as of dates as per every month end run, then selected frequency for this field must be at least monthly or more than this, could be quarterly, Semi-Annually or Annually. This is required as there are models in subsequent steps like Drawdown Analysis, GBM or SARIMAX which needs series of data points for given as of date range.

Moving Average Period – This defines the period over which moving average gets calculated to create Bollinger band. Sigma Co-efficient helps to define the lower and upper bands for Bollinger band graphs.

Pass-Through rate Calc

Pass Through Rate is based on linear regression between market and offered rate. For current accounts, offered rate can be zero, in those scenarios, pass through rate will be zero, as any change in market rate doesn’t have any impact on offered rate.

Stable Balance Calculations

Drawdown Analysis: This model does calculations based on maximum drawdown between two data points for given period. Say analysis is done over past 5 years and monthly data points are given. On each data point, balance is available, system calculates balance reduction for each data point from previous data point. Finally, whatever will be the maximum balance reduction in two adjacent data points that is subtracted from point in time balance (Balance on latest as of Date) and remaining balance is the stable balance.

Geometric Brownian Motion – This model works based on historical volatility and confidence interval selected in the UI. It assumes underlying data follows normal distribution, and as per the confidence interval, X% of the balance falls between below ranges; which helps the system derive stable balance.

95% Confidence Interval – (-1.96 to +1.96)

90% Confidence Interval – (-1.645 to +1.645)

80% Confidence Interval – (-1.282 to +1.282)

Here User has the option to choose outlook scenario factor, which will be multiplied with historical volatility and subsequently used for GBM based stable balance calculations. GBM outcome is based on both portfolio level balance volatility and account level balance volatility. User can select any of the stable balances, which are more suitable as per the given context.

Decay Rate Calculations

Vintage Analysis: The vintage run-off model seeks to segregate deposit balances based on historical tranches called vintages. A vintage consists of all of the individual accounts for a non-maturing deposit account type that are opened within a particular time bucket; it could be month, week. Behavioral characteristics are applied to each vintage by calculating the monthly decay within each vintage. Run off calculations are done across data points and across account origination buckets and finally an average is taken across both dimensions to arrive at final decay rate for the selected portfolio.

Vintage analysis mandatorily needs account level data to segregate accounts by their origination dates, in case account level data is not available, there is a checkbox given in UI, which can skip this model.

SARIMAX: Among the various approaches available, the SARIMAX (Seasonal Autoregressive Integrated Moving Average + exogenous variables) model stands out as a powerful tool for modeling and forecasting both trends and seasonal variations in temporal data, while incorporating exogenous variables into the analysis to improve prediction accuracy. This model belongs to ARIMA models.

It is based on three key components: autoregression (AR), moving average (MA) and integration (I).

Autoregression (AR) considers past values of the time series to predict current values. The moving average (MA), on the other hand, tackles past errors in predictions. The moving average consists of performing a linear regression on the last q error values to predict the current value.

When temporal data show seasonal variations, the SARIMA model takes over the scene. The term “Seasonal” is added to ARIMA to indicate that this model can capture patterns that repeat at regular intervals. Based on the decay rate in each of the periods, weighted average life is calculated by multiplying run-offs matrix by time series.

GBM (Geometric Brownian Motion): Conceptually, this is same as mentioned above under stable balance, it is extended here to capture forecasted balance and how those deteriorate over forecasted period. Using regular run-offs in each period and multiplying those with period itself, weighted average life is calculated.

If there is any regulatory cap on WAL of non-maturity accounts like CASA, then system automatically perform iteration by changing confidence interval to keep WAL under the given regulatory cap, and accordingly produces decay rate profile as well.