Quantum Computing for Finance — Final Exam

This is the graded finale for Quantum Computing for Finance, and it runs across the whole sober map. You started with the foundations: a qubit holds a superposition $a\lvert 0\rangle + b\lvert 1\rangle$ with $\lvert a\rvert^2 + \lvert b\rvert^2 = 1$, measurement collapses it, and the ”$2^n$ states at once” headline buys you nothing unless interference concentrates amplitude on the answer you want — and you learned the two hardware worlds, the universal gate model (QAE, QAOA) versus special-purpose quantum annealing (Ising/QUBO). Then the crown jewel: amplitude estimation turns classical Monte Carlo’s $1/\sqrt{N}$ error into $1/N$, a genuine quadratic speedup for option pricing and VaR/CVaR — paid for by an oracle you must assume is cheap. You recast portfolio optimization as QUBO ($x^\top Q x$ over binary $x$), enforced cardinality with a penalty $\lambda(\sum_i x_i - k)^2$, and solved it with QAOA or annealing — strong mainly where discrete constraints make classical convex solvers choke. You audited quantum machine learning — quantum kernels and variational classifiers — with the deepest skepticism, since expressiveness was never finance’s binding constraint. And you walked the walls: NISQ noise that kills circuit depth, error-correction overhead of ~1000+ physical qubits per logical one, and above all the data-loading problem that can erase the very speedup it serves. The payoff: quantum-inspired classical methods like tensor networks that run today, judged — like everything here — end-to-end against the best classical baseline. No hints are shown, each answer locks the instant you submit, and your score stays hidden until the very end.

Course Recap