By the end you will be able to
- 1Backtest each kind of behavioural assumption: rate pass-through, core balances, stressed run-off and prepayment.
- 2Explain why out-of-sample windows and the choice of what is tested decide whether a backtest can fail at all.
- 3Test a prepayment model's bias with a t statistic and confidence interval, and judge whether run-off breaches are independent.
- 4Keep an assumption register with owners, versions and traffic lights, and say what a red light should trigger.
Why it matters
Every figure this module produced rests on a short list of assumptions: how fast deposit rates follow Bank Rate, how much of each balance stays, how much could run in a month, how fast borrowers prepay. In March 2023 RiskAtoms Bank's deposit books each lost more in a month than its 99% run-off model allowed. Yet the Risk Engine's register, on its default window, shows every core balance green. Both are right. What a backtest finds depends on what it tests, against what, and over which months.
Module complete