Market, credit, liquidity and model-risk reasoning
Financial risk concerns uncertain outcomes that affect objectives, resources or obligations. Define the exposure, horizon, unit of measurement and decision before choosing a metric. Market risk arises from changing prices, rates, spreads or exchange values; credit risk concerns default or deterioration; liquidity risk concerns meeting obligations or trading without excessive cost. Operational risk can arise from people, systems, processes and external events. These categories can interact, so a label is a starting point rather than a complete causal explanation.
A simple credit expected-loss model multiplies probability of default, loss given default and exposure at default when the definitions and horizon are consistent. In a fictional one-year example, probability2percent, exposure1000and loss severity40percent give expected loss8. That average is not the maximum possible loss and does not establish the applicable accounting allowance or regulatory capital requirement. Recovery timing, collateral, dependence, changing exposures and economic conditions can require a richer model. Distinguish an expected loss from unexpected variation around it.
Value at Risk describes a loss quantile for a stated model, confidence level and horizon. A one-day99percent VaR of10does not say losses can never exceed10or specify how large tail losses might be. Expected shortfall examines losses beyond a specified quantile under its definition. Estimates depend on data, distributional assumptions, dependence and valuation methods. Historical observations do not guarantee that future extremes follow the same pattern. Stress tests explore specified adverse scenarios, which need not have assigned probabilities.
Liquidity analysis separates funding needs from the ability to sell an asset. Contractual and behavioral cash-flow timing can differ. An institution may face collateral calls, deposit withdrawals or unavailable refinancing at the same time that market prices weaken. Buffer assets can be encumbered, less liquid than expected or exposed to concentration. Avoid inferring resilience from a single balance-sheet ratio or applying an unsupported universal threshold. Scenario design should consider combined stresses and the practical feasibility of proposed responses.
Model governance includes purpose, data provenance, assumptions, validation, monitoring and accountable use. Backtesting compares predictions with outcomes using an appropriate rule, but passing one historical test does not establish permanent reliability. Training and test leakage can make a model appear more predictive than it is. A regression association is not automatically causal, and a fitted relationship may fail outside the observed range. Document limitations and evaluate independent evidence. Risk transfer can change who bears a loss without eliminating all residual, legal or counterparty exposure. These are educational risk foundations, not an assertion of current Basel, solvency or national investment requirements.