Programming

Numerical Methods for Quant Interviews

NeetQuant · August 2026 · 4 min read

Root finding

Used constantly to back out implied volatility from an option price.

Bisection - always converges if the root is bracketed, but slowly (linear). Robust.

Newton-Raphson - quadratic convergence, so it roughly doubles the correct digits each step. But it needs the derivative and a decent starting point, and it can diverge.

For implied vol the derivative is vega, which is available analytically, so Newton is the standard choice - usually with a bisection fallback for robustness. Being able to say why you would combine them is a strong answer.

Monte Carlo

Simulate many paths, average the payoffs.

The key fact: standard error falls like 1/sqrt(N). Four times the samples for half the error - the same square-root law as everywhere else.

Crucially, this rate is independent of dimension, which is why Monte Carlo beats grid methods for high-dimensional problems even though it is slow in one dimension.

Variance reduction

More valuable than more samples, and a good thing to raise unprompted:

  • Antithetic variates - use each random draw and its negative, exploiting symmetry.
  • Control variates - subtract a correlated quantity whose exact value you know.
  • Importance sampling - sample more where the payoff is non-zero, essential for deep out-of-the-money options where almost all paths contribute nothing.

Floating point

Interviewers do test this.

  • Never use exact equality. Compare with a tolerance.
  • 0.1 + 0.2 does not equal 0.3 in binary floating point.
  • Catastrophic cancellation: subtracting two nearly equal numbers destroys precision. The classic case is the quadratic formula, where one root should be computed via the alternative form.
  • Summing many small numbers into a large accumulator loses precision; Kahan summation fixes it.

Numerical differentiation

Finite differences for Greeks when no analytic form exists. Central differences are second-order accurate and preferable to forward differences. The step size trades truncation error against floating-point noise, with an optimum around the square root of machine epsilon.

Practise in programming and DSA.

Keep practising

Practise quant interview questions free

Create a free account to attempt hundreds of questions with hints and answer checking, and to run the timed simulators.

Start practising free

Frequently asked questions

How is implied volatility computed numerically?
By root finding on the pricing function, usually Newton-Raphson using vega as the derivative for fast convergence, with a bisection fallback for robustness when Newton diverges.
How does Monte Carlo error scale?
Standard error falls like 1/sqrt(N), so four times the samples halves the error. The rate is independent of dimension, which is why Monte Carlo wins for high-dimensional problems.