Korea Financial Risk Manager common mistakes
Korea Financial Risk Manager · Understand the traps and how to avoid them.
Memorizing the variance-covariance VaR formula but applying it to a portfolio without converting individual asset volatilities and correlations into a covariance matrix, then summing standalone VaRs instead of taking the square root of the quadratic form.
Candidates drill single-asset VaR until it is automatic, then treat multi-asset VaR as simple addition because the arithmetic feels familiar; the sub-additivity of diversified VaR is counterintuitive under exam time pressure.
What to do instead
Before computing, write the portfolio variance expression explicitly: w'Σw with correlations off the diagonal. Practice at least ten two- and three-asset cases where the undiversified sum and the diversified VaR differ materially, and state in one line why the diversified figure is lower.
Treating the delta-normal approach as valid for option portfolios with large moves, reporting a VaR that ignores gamma and vega, and then failing follow-up items that ask for the delta-gamma or full-revaluation adjustment.
Delta-normal is the first method taught and the easiest to compute, so candidates default to it without checking whether the position is linear or whether the shock is large relative to the option's convexity.
What to do instead
Build a decision rule: if the portfolio contains options and the confidence level or horizon implies a move beyond roughly one standard deviation, flag convexity. Memorize the delta-gamma correction term (one-half times gamma times the squared shock) and be able to explain when full revaluation is required.
Mixing up expected shortfall and VaR in coherence questions, claiming VaR is coherent, or computing ES as a simple average of VaR and the tail loss rather than the conditional mean of losses beyond the VaR threshold.
The two measures are introduced together and both are tail statistics, so the definitions blur; the coherence axioms (monotonicity, translation invariance, positive homogeneity, sub-additivity) are rarely applied by hand.
What to do instead
Write the ES definition as E[L | L > VaR] and verify sub-additivity with a two-position numerical example where VaR violates it. Keep a one-page table mapping each axiom to a pass or fail for VaR and ES, and re-derive it from scratch rather than recalling it.
Applying the wrong scaling rule to VaR, such as multiplying a one-day 99% VaR by the square root of 250 to get an annual figure, or scaling a 95% VaR to 99% by multiplying by a constant without using the normal quantile ratio.
The square-root-of-time rule is taught as a universal shortcut, and candidates forget it assumes i.i.d. returns and applies to volatility, not directly to a quantile unless returns are normal.
What to do instead
Distinguish scaling of volatility (multiply by square root of time) from scaling of a quantile (multiply by the ratio of normal quantiles, e.g. 2.326/1.645 for 99% versus 95%). Note explicitly that the rule fails under autocorrelation or volatility clustering and state the direction of the resulting bias.
Confusing the probability of default, exposure at default, and loss given default when computing expected loss, for example multiplying PD by EAD by recovery rate instead of by one minus recovery rate, or using a cumulative multi-year PD where a marginal PD is required.
The three credit parameters appear in several formulas (expected loss, unexpected loss, credit VaR, CVA) and the recovery-rate convention flips between LGD and recovery depending on the source, so sign and complement errors creep in.
What to do instead
Always write EL = PD x EAD x LGD with LGD = 1 - recovery, then substitute numbers. For multi-year problems, convert cumulative PD to marginal PD using the survival-probability relationship before applying it to each period.
Reading a credit rating transition matrix as if each row were unconditional default probabilities, or failing to compound the one-year default probability over multiple years using (1 - PD)^n for survival rather than multiplying PD by n.
Transition matrices are usually presented with default in the last column, and candidates skim the row total instead of the default cell; multi-year survival is frequently approximated linearly under time pressure.
What to do instead
Practice extracting the default column explicitly and computing n-year survival as (1 - PD)^n under the constant-hazard assumption, then converting to cumulative PD as 1 minus that quantity. Compare the linear approximation to the exact figure to see the size of the error.
Computing the Merton model distance to default with the wrong numerator, omitting the drift term or subtracting the risk-free rate incorrectly, and then misinterpreting a higher distance to default as higher risk.
The structural model combines asset value, debt face value, asset volatility, and drift in one expression, and the sign convention for distance to default is easy to invert.
What to do instead
Write DD = (ln(V/D) + (mu - 0.5 sigma^2)T) / (sigma sqrt(T)) and state in words that DD is the number of standard deviations of asset value above the default point, so larger DD means lower default probability. Sanity-check with a numerical example.
Treating the binomial tree and Black-Scholes as interchangeable without checking whether early exercise matters, and pricing an American put with the Black-Scholes formula.
Black-Scholes is the default tool in candidates' minds, and the no-dividend European assumption is easy to overlook when the question does not state it prominently.
What to do instead
Before pricing, classify the option: American versus European, dividend-paying versus not. Use the tree for American options and for discrete dividends, and remember that early exercise of an American call without dividends is never optimal while an American put can be exercised early.
Miscalculating duration and convexity effects by using modified duration where Macaulay duration belongs, or applying the convexity adjustment with the wrong factor of one-half and the wrong units for the yield change.
The three duration measures differ by small yield-based adjustments, and the second-order Taylor expansion is memorized as a formula without tracking whether the yield change is in decimal or percentage-point form.
What to do instead
Write the full price-change approximation as -D_mod times delta-y plus one-half times convexity times delta-y squared, with delta-y in decimal. Convert Macaulay to modified duration by dividing by one plus yield over the compounding frequency, and test with a 100 basis point move to confirm the sign and magnitude.
Answering governance and risk-appetite questions with generic statements about 'strong risk culture' instead of naming the specific mechanism, such as the three-lines-of-defense model, board-level risk committee responsibilities, or the link between risk appetite and limit-setting.
Foundations material reads as qualitative, so candidates skim it and rely on intuition, but exam items test precise allocation of responsibility and the distinction between risk appetite, tolerance, and limits.
What to do instead
Build a one-page map of roles: board sets appetite, risk committee oversees, first line owns risk, second line sets frameworks, third line audits. For each, memorize one concrete artifact (appetite statement, limit ladder, escalation trigger) and practice writing it in a single sentence.