ALM Analytics, Forecasting and Optimization

Introduction and Context

Asset–liability management (ALM) converts a bank’s strategic objectives and risk appetite into decisions about balance-sheet composition, funding, liquidity, pricing and capital. Earlier ALM practice relied heavily on maturity ladders, gap reports and expert judgement. Modern treasury functions combine these tools with behavioural modelling, statistical forecasting, scenario analysis, simulation and constrained optimization.

The central analytical problem is not simply to forecast interest rates or deposits. It is to understand how uncertain cash flows, customer behaviour, market prices, funding access, accounting effects and regulation interact. A useful ALM model therefore has four characteristics: it is economically interpretable, internally consistent, stress-capable and governed through independent validation.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Construct contractual and behavioural cash-flow projections.
2. Distinguish deterministic forecasts, scenarios, simulations and optimization.
3. Estimate behavioural assumptions for non-maturity deposits, prepayments and drawdowns.
4. Model liquidity, net interest income (NII), economic value and funding concentration.
5. Formulate a balance-sheet optimization problem subject to regulatory and managerial constraints.
6. Evaluate model risk, governance, fairness and implementation issues.

1. The ALM Analytical Architecture

An ALM analytics process normally follows this sequence:

Data → behavioural assumptions → cash-flow engine → market scenarios → risk measures → optimization or decision → limits and monitoring.

The cash-flow engine maps each asset, liability and off-balance-sheet position into time buckets. For instrument i and period t, projected net cash flow may be represented as:

All terms must use the same currency and time interval. A positive value is a net inflow. Portfolio cash flow is:

A cash-flow forecast is not identical to an accounting forecast. Cash flows describe timing of receipts and payments; NII describes accrued interest income and expense; economic value reflects discounted present values; liquidity metrics may apply regulatory definitions and stress assumptions.

Contractual versus behavioural cash flows

Contractual cash flows are specified by legal terms: maturity dates, amortization schedules, repricing dates, coupon rules and contractual options. Behavioural cash flows incorporate observed or assumed customer actions, such as early repayment, deposit attrition, overdraft utilization or renewal of maturing wholesale funding.

A contractual maturity ladder is essential but incomplete. A five-year mortgage may be contractually outstanding for five years while historically prepaying much earlier. A demand deposit may have no contractual maturity but may be operationally stable. Behavioural modelling transforms such positions into expected, rather than merely legal, cash flows.

2. Forecasting Cash Flows and Balance-Sheet Drivers

2.1 Building a projection model

A robust projection model should contain, at minimum:

• instrument-level balances and contractual terms;
• repricing, maturity and amortization dates;
• reference rates, spreads and floors;
• expected new business and run-off;
• behavioural assumptions;
• funding and collateral rules;
• accounting and regulatory mappings; and
• data quality and version controls.

For an amortizing loan with opening principal P₀, periodic interest rate r, and level payment A over n periods, the payment under a fully amortizing fixed-rate structure is:

This formula assumes equal periodic payments, a constant rate, no fees, no prepayments and payments at the end of each period. If rates reset or prepayments occur, the contractual schedule must be recalculated rather than applying this formula mechanically.

For a simple cash-flow forecast, expected balance at the next period can be expressed as:

The terms must be defined consistently. For example, a charge-off reduces principal but may not be a cash outflow at the same time as recognition.

2.2 Roll-rate and cohort methods

A roll-rate model tracks movements between delinquency or product states. If x_t is a vector of balances across states and P is a transition matrix, then:

Here o_t represents new originations and r_t removals such as repayments or charge-offs. Each column of P should sum to one when it represents mutually exclusive exhaustive transitions. The transition probabilities may depend on borrower characteristics, unemployment, interest rates or loan age.

Cohort analysis groups accounts by origination month, product, rate type or customer segment. It is especially useful for identifying seasoning effects, vintage deterioration and changes in underwriting quality.

2.3 Forecasting methods

No single forecasting method is universally superior. Common approaches include:

A basic regression forecast can be written as:

where y_t is the forecast variable, x_j,t are explanatory variables, β_j are estimated coefficients and ε_t is an error term. The relationship should not be interpreted causally unless the design supports causal inference. Forecast errors should be assessed for bias, autocorrelation, instability and performance during stress.

3. Behavioural Assumptions

3.1 Non-maturity deposits

Demand deposits, current accounts and many savings products have no contractual maturity. ALM commonly estimates an effective maturity using a combination of:

• deposit decay or attrition;
• core and volatile balance segmentation;
• sensitivity to market rates;
• customer and product characteristics;
• operational stickiness; and
• observed rate pass-through.

A simple exponential decay model is:

where B₀ is the initial balance, B_t is the expected balance after time t, and λ is the continuous attrition rate per unit of time. The model assumes a constant decay rate and is therefore a simplifying representation, not a universal law. A discrete monthly decay model is:

where d is the monthly attrition rate. These models should be segmented and tested against actual runoff. A deposit balance that is stable in normal conditions may become highly rate-sensitive or flight-prone during stress.

Deposit beta measures pass-through from a market-rate change to a deposit-rate change. For a product over a selected observation period:

The ratio is meaningful only when the rate changes are comparable, the observation window is specified and the reference rate is economically relevant. It may be asymmetric: pass-through during rate increases need not equal pass-through during rate decreases.

3.2 Prepayment and early redemption

For loans, prepayment assumptions affect both liquidity and interest-rate risk. A conditional prepayment rate (CPR) is an annualized rate; its equivalent single monthly mortality (SMM), under a standard conversion, is:

If the scheduled balance subject to unscheduled repayment is S_t, expected unscheduled repayment in the month is approximately S_t × SMM, subject to product-specific conventions. The conversion assumes a constant monthly rate and excludes complications such as seasoning, burnout and refinancing incentives.

Prepayment models should consider the borrower’s refinancing incentive, seasoning, burnout, borrower credit quality, penalties, housing turnover and operational frictions. A model based only on the current rate gap can overstate refinancing in periods when credit or capacity constraints are binding.

3.3 Credit-line utilization and contingent liquidity

Undrawn commitments create potential future cash outflows. If undrawn commitment is U_t and expected utilization rate is q_t, expected draw is:

For stress analysis, q_t may be increased using observed crisis utilization, segment-specific estimates or a prescribed regulatory factor. The bank must avoid double-counting: a draw that becomes a funded loan should be treated as a cash outflow initially and as an asset thereafter, with subsequent interest and repayment flows modelled separately.

4. Scenario Analysis and Simulation

4.1 Deterministic scenarios

A deterministic scenario specifies a coherent path for variables such as policy rates, yield curves, exchange rates, unemployment, deposit attrition and credit spreads. Examples include parallel rate shocks, steepening or flattening curves, rapid inflation, recession, market closure and idiosyncratic deposit outflow.

A scenario is not merely a collection of independent shocks. Economic consistency matters. For example, a recession may reduce loan demand, increase defaults, increase drawdowns and alter deposit behaviour simultaneously.

4.2 Sensitivity, stress and reverse stress testing

A sensitivity changes one or a small number of inputs to identify exposure. Stress testing applies a severe but plausible combination of changes. Reverse stress testing begins with a failure outcome—such as breaching a survival horizon or minimum capital ratio—and identifies combinations of events capable of producing it.

Scenario design should include historical, hypothetical, regulatory and institution-specific scenarios. Results should be communicated through both absolute outcomes and changes from the base case.

4.3 Monte Carlo simulation

Simulation represents uncertain drivers by probability distributions and generates many possible paths. If a risk factor follows a geometric Brownian-motion approximation:

where S_t is the risk-factor level, μ is drift, σ is volatility and dW_t is a Wiener increment. This model is not a claim that rates or deposits literally follow geometric Brownian motion; it is an assumption that must be justified. Short-rate, yield-curve, deposit and credit models often require different structures.

For simulated outcomes Z₁,…,Z_N, the estimated mean is:

The Monte Carlo standard error of the mean is approximately s/√N, where s is the sample standard deviation. More simulations reduce sampling error but do not correct model misspecification, poor correlations or unrealistic distributions.

Correlation is especially important. A deposit outflow, widening credit spread and collateral haircut may occur together. Using independent random draws can substantially understate joint liquidity risk.

5. ALM Risk Measures

5.1 Liquidity cash-flow metrics

A cumulative liquidity position over horizon T can be expressed as:

This is a management measure whose definitions must specify eligible assets, haircuts, operational restrictions and stress assumptions. Regulatory liquidity ratios such as the Liquidity Coverage Ratio (LCR) and Net Stable Funding Ratio (NSFR) use detailed prescribed definitions and should not be recreated from simplified equations for reporting purposes.

A useful internal metric is the survival horizon: the first period in which cumulative stressed liquidity becomes negative after applying usable liquidity reserves and realistic monetization constraints.

5.2 Net interest income

For a period, simplified NII is:

A more explicit approximation is:

where asset balances A_a,t and liability balances L_l,t are measured consistently, and rates r correspond to the period convention. This is a simplification: day-count rules, average balances, non-accrual assets, transfer pricing, hedges and accounting classifications may materially affect reported NII.

5.3 Economic value sensitivity

The present value of cash flows is:

For continuously compounded discounting, the denominator is replaced by e^y_tt under the relevant convention. y_t must be the discount rate applicable to the cash flow’s currency, tenor and risk characteristics.

Economic value of equity (EVE) is commonly represented as:

An interest-rate shock produces ∆EVE = EVE_shock − EVE_base. EVE is long-term and market-value oriented; NII is shorter-term and earnings oriented. A strategy may improve one while weakening the other.

Duration provides a first-order approximation for a small parallel yield change:

Here D_M is modified duration in years and ∆y is the decimal yield change. The approximation is unreliable for large shocks, significant convexity, embedded options or nonparallel curve movements.

6. Optimization of Balance-Sheet Composition

Optimization formalizes a choice among feasible strategies. Let x be a vector of decision variables, such as loan growth, securities allocation, deposit pricing, wholesale funding, hedges and liquid-asset reserves. A generic problem is:

subject to:

Constraints may include minimum liquidity, maximum concentration, capital requirements, leverage, NSFR, risk appetite, funding capacity, limits on duration or basis risk, and operational capacity. α and β are management risk-preference weights; they should be approved, documented and subjected to sensitivity analysis.

A multi-scenario formulation can protect against adverse outcomes:

where p_s are scenario weights satisfying p_s ≥ 0 and ∑p_s = 1. Scenario probabilities are often uncertain, so expected-value optimization should be supplemented by stress constraints or robust optimization.

A robust formulation may maximize the worst-case objective over an uncertainty set Ω:

This can be conservative and may reduce profitability. Its value lies in making uncertainty explicit rather than disguising it as false precision.

Linear, nonlinear and stochastic optimization

Linear programming is suitable when objective functions and constraints are linear. Quadratic programming can represent variance or tracking-error penalties. Nonlinear optimization is needed for options, nonlinear pricing and convexity. Stochastic programming represents decisions at multiple dates and information states. The selected method should reflect model structure, data quality and explainability, not mathematical sophistication alone.

Shadow prices from constrained optimization can be managerially informative. A binding liquidity constraint may reveal the marginal value of additional liquid assets; however, the interpretation is valid only under the model’s assumptions and local changes.

7. Worked Example: Funding-Mix Decision

Consider an illustrative bank choosing between two funding sources for a one-year requirement of 100 monetary units. Stable deposits cost 3.0% and wholesale funding costs 4.2%. Management requires at least 40 units of stable funding, limits wholesale funding to 60 units and assigns a liquidity-risk penalty of 0.8% to wholesale funding. Let x be stable deposits and 100−x wholesale funding.

The annual modeled cost is:

Since the effective wholesale cost is 5.0%,

The cost decreases as x increases. The constraints are 40 ≤ x ≤ 100, but an internal policy may also require at least 20 units of wholesale diversification. If that policy is imposed, x ≤ 80, and the optimum is x = 80. Cost is:

Thus the selected mix is 80 units of stable deposits and 20 units of wholesale funding, with cost 3.4 monetary units. The example demonstrates that the cheapest unconstrained funding source is not automatically the optimal funding strategy. Diversification, capacity and liquidity risk can create binding constraints.

8. Applied Case: Illustrative Regional Bank

Background

Riverland Bank, a fictional regional bank, has 70% floating-rate commercial loans, a growing mortgage portfolio, substantial non-maturity deposits and moderate reliance on three-month wholesale funding. Its board has approved risk appetite limits for a 30-day stressed liquidity gap, one-year NII sensitivity and EVE sensitivity.

Problem

The treasury team’s contractual report shows a comfortable liquidity position. However, deposits have become more rate-sensitive, mortgage prepayments have increased and undrawn corporate facilities were heavily utilized during a recent market disruption. The contractual report therefore understates plausible stress outflows.

Analysis

The bank segments deposits into operational, relationship and rate-sensitive balances. It estimates separate decay and deposit-beta parameters, validates them on monthly observations and applies a severe but plausible attrition scenario. Mortgage cash flows are projected with a prepayment model incorporating refinancing incentive and seasoning. Corporate commitment utilization is increased using crisis observations.

Three scenarios are constructed:

The cash-flow engine produces daily liquidity projections and monthly NII and EVE results. An optimization model chooses deposit pricing, liquid securities, wholesale tenor and loan growth subject to liquidity, capital, concentration and duration constraints.

Approach and solution

The recommended strategy is not to maximize loan growth. It includes a larger immediately monetizable liquidity buffer, a longer wholesale maturity profile, segmented deposit pricing rather than blanket rate increases, and a cap on growth in products with high contingent liquidity usage. The optimization is rerun under parameter ranges rather than a single point estimate.

Outcome

In the illustrative result, base-case NII declines slightly because of the liquidity buffer and higher funding tenor cost. However, the stressed survival horizon improves materially, wholesale refinancing concentration falls and the worst-case EVE loss remains within risk appetite. The board approves the strategy with quarterly assumption review and monthly early-warning monitoring.

Learning points

1. Contractual maturity is not a sufficient liquidity forecast.
2. Behavioural assumptions can dominate reported risk measures.
3. Optimization must include constraints and stress outcomes.
4. A profitable base case can be unacceptable if it fails liquidity stress.
5. Model outputs are decision support, not substitutes for treasury judgement.

9. Governance, Regulation and Model Risk

ALM models require clear ownership, documented purpose, controlled data, independent validation and disciplined change management. Validation should examine conceptual soundness, data lineage, implementation accuracy, performance monitoring, sensitivity, benchmarking and limitations.

Important controls include:

• reconciliation of modeled balances to the general ledger and risk systems;
• independent review of assumptions and scenario design;
• back-testing of forecasts against realized runoff and repricing;
• challenger models and expert overlays;
• version control and approval evidence;
• explicit treatment of model uncertainty; and
• escalation when limits or data-quality thresholds are breached.

Regulatory requirements differ by jurisdiction, but common supervisory themes include liquidity adequacy, interest-rate risk in the banking book, stress testing, governance, capital and prudent valuation. A model should never be presented as regulatory-compliant merely because it produces a familiar metric; regulatory definitions, reporting rules and supervisory guidance must be applied in context.

Ethically, behavioural models may influence deposit pricing, credit availability and customer treatment. Segmentation should not create unlawful discrimination or exploit vulnerable customers. AI models require explainability proportional to impact, monitoring for drift and bias, controls over alternative data, privacy protection and human accountability. Generative AI may assist documentation or code review, but unverified generated logic must not enter a production ALM engine.

10. Contemporary Developments and Best Practices

Contemporary ALM is moving toward integrated data platforms, intraday liquidity monitoring, high-frequency payment information, cloud-scale simulation, automated model monitoring and scenario libraries. Digital deposits can move rapidly, making historical “stickiness” less reliable. Climate risk may affect collateral values, insurance, funding access and long-term cash flows. Tokenized or instant-settlement instruments may alter payment timing and liquidity demand.

Best practice is to:

1. Begin with economic purpose and decision use, not algorithm selection.
2. Separate contractual facts from behavioural assumptions.
3. Use transparent assumptions with confidence ranges.
4. Combine base forecasts, sensitivities, severe stresses and reverse stresses.
5. Model correlations and feedback effects where material.
6. Optimize only within clearly defined feasible and governed constraints.
7. Report both expected performance and downside resilience.
8. Maintain contingency funding actions linked to early-warning indicators.
9. Validate models independently and monitor drift continuously.
10. Preserve management override, while recording its rationale and outcome.

Challenges and Limitations

ALM analytics faces structural uncertainty, limited crisis data, regime changes, feedback between bank actions and customer behaviour, parameter instability, data lags and conflicting objectives. A deposit model calibrated during a low-rate period may fail after a rapid tightening cycle. Optimization may produce a mathematically precise answer that is operationally infeasible. Scenario probabilities may be unreliable, and stress losses may be nonlinearly amplified by collateral, margin and reputation effects.

The appropriate response is not to abandon quantitative methods. It is to use them with ranges, challenger views, qualitative judgement, conservative overlays where justified and transparent communication of uncertainty.

Chapter Summary

ALM analytics integrates cash-flow projection, behavioural modelling, scenario analysis, simulation and optimization. Contractual schedules provide the legal baseline, while deposit decay, deposit beta, prepayment, drawdown and attrition models estimate economically plausible behaviour. Forecasts should be evaluated for bias, instability and stress performance.

Liquidity, NII and EVE measure different dimensions of risk. Deterministic scenarios explain sensitivities; Monte Carlo simulation explores distributions and joint uncertainty; reverse stress testing identifies pathways to failure. Optimization can select funding, asset and hedge strategies, but only subject to liquidity, capital, concentration, regulatory and operational constraints. Governance is essential because model error, data weakness and behavioural change can dominate numerical precision. Effective ALM therefore combines quantitative discipline with judgement, independent validation and clear accountability.

Key Terms

Asset–liability management: Coordinated management of assets, liabilities, liquidity, interest-rate risk and capital.

Behavioural cash flow: Expected cash flow based on customer or market behaviour rather than legal terms alone.

Deposit beta: Change in deposit rate divided by change in a reference market rate.

Decay rate: Estimated rate at which a balance runs off over time.

Prepayment: Repayment before contractual maturity.

Conditional prepayment rate: Annualized measure of expected prepayment.

Single monthly mortality: Monthly equivalent of an annual prepayment rate.

Roll-rate model: Model of movements between portfolio states.

Scenario analysis: Evaluation under specified coherent assumptions.

Stress testing: Analysis of severe but plausible adverse conditions.

Reverse stress testing: Identification of conditions that could cause failure or limit breach.

Monte Carlo simulation: Repeated random generation of outcomes from specified distributions.

Net interest income: Interest income less interest expense, subject to accounting conventions.

Economic value of equity: Present value of assets less present value of liabilities.

Duration: First-order sensitivity of present value to yield changes.

Optimization: Selection of the best feasible decision under an objective and constraints.

Scenario probability: Weight assigned to a modeled scenario.

Model risk: Risk of adverse consequences from incorrect model design, implementation or use.

Review Questions

Short Answer Questions

1. Why are contractual cash flows insufficient for ALM decision-making?
2. Define deposit beta and identify one reason it may be asymmetric.
3. Distinguish sensitivity analysis from stress testing.
4. What is the purpose of a reverse stress test?
5. Why must an optimization model include operational constraints?

Long Answer Questions

1. Explain how a bank can forecast non-maturity deposits using decay, segmentation and deposit-beta analysis. Discuss limitations.
2. Compare NII sensitivity and EVE sensitivity as measures of interest-rate risk.
3. Describe the architecture of a bank-wide ALM cash-flow engine.
4. Explain how Monte Carlo simulation can support liquidity risk analysis, including the importance of correlations.
5. Discuss governance requirements for behavioural ALM models and the role of independent validation.

Analytical / Application Questions

1. A bank has 200 monetary units of deposits and assumes a monthly decay rate of 1.5%. Calculate the expected balance after 12 months using B_t = B₀(1−d)^t, and discuss why the result may not hold during a crisis.
2. Construct a two-scenario funding optimization for a bank choosing stable deposits, wholesale funding and liquid assets. Specify at least four constraints and explain the trade-offs.
3. A mortgage portfolio shows increased prepayments after market rates fall, while deposits become more rate-sensitive. Design a coherent scenario and explain its effects on liquidity, NII and EVE.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes contractual and behavioural cash flows?

A. Contractual cash flows are always more conservative than behavioural cash flows.

B. Behavioural cash flows incorporate expected customer actions in addition to legal terms.

C. Behavioural cash flows are used only for accounting purposes.

D. Contractual cash flows require probability distributions.

Correct Answer: B. Behavioural cash flows incorporate expected customer actions in addition to legal terms.

Question 2

If a deposit balance is modeled as B_t = B₀(1−d)^t, what does d represent?

A. The annual interest rate

B. The discount factor

C. The periodic attrition rate

D. The deposit beta

Correct Answer: C. The periodic attrition rate

Question 3

What is the principal purpose of reverse stress testing?

A. To estimate the most likely base-case outcome

B. To identify conditions that could cause a specified failure or limit breach

C. To remove all model uncertainty

D. To calculate contractual loan amortization

Correct Answer: B. To identify conditions that could cause a specified failure or limit breach

Question 4

In the simplified expression NII = Interest income − Interest expense, NII primarily measures:

A. Present value of equity

B. Net interest earnings over a period

C. Liquidity survival time

D. Capital adequacy

Correct Answer: B. Net interest earnings over a period

Question 5

Which condition is essential for scenario probabilities in an expected-value model?

A. Every probability must exceed 0.5

B. Probabilities must be negative for stress scenarios

C. Probabilities must be nonnegative and sum to one

D. All scenarios must have identical outcomes

Correct Answer: C. Probabilities must be nonnegative and sum to one

Question 6

A bank has 100 units of undrawn commitments and an expected utilization rate of 25%. What is the expected draw?

A. 4 units

B. 25 units

C. 75 units

D. 125 units

Correct Answer: B. 25 units

Question 7

The first-order duration approximation ∆PV ≈ −D_M × PV × ∆y is most appropriate when:

A. The yield change is small and the instrument is not strongly nonlinear

B. The instrument has unlimited prepayment options

C. The yield curve changes by an unknown amount

D. Cash flows are not discounted

Correct Answer: A. The yield change is small and the instrument is not strongly nonlinear

Question 8

Why can independent simulation of deposit outflows and collateral haircuts understate liquidity risk?

A. It guarantees both variables remain constant.

B. It ignores possible joint occurrence and correlation.

C. It always overstates volatility.

D. It converts liabilities into equity.

Correct Answer: B. It ignores possible joint occurrence and correlation.

Question 9

In an ALM optimization model, a binding liquidity constraint may indicate:

A. The marginal value of relaxing that constraint, subject to model assumptions

B. That liquidity has no economic value

C. That all other constraints are irrelevant

D. That the model requires no validation

Correct Answer: A. The marginal value of relaxing that constraint, subject to model assumptions

Question 10

Which is an appropriate governance practice for a behavioural ALM model?

A. Change assumptions without recording revisions

B. Use only one historical period regardless of regime

C. Perform independent validation and monitor forecast performance

D. Allow generated code into production without testing

Correct Answer: C. Perform independent validation and monitor forecast performance