Quantum Computing for Finance

Quantum computing has spent a decade as finance’s favorite shimmering mirage: every bank press release, every conference keynote, every “we’re quantum-ready” white paper promising that exponential speedups are about to vaporize your option-pricing grid and solve portfolio optimization in the blink of an eye. Then you look for the production system actually trading on a quantum computer, and you find — a research pilot, a 12-qubit toy problem, and a footnote admitting the classical baseline still wins.

This course is the antidote to both the hype and the reflexive dismissal. Quantum computing is not magic finance pixie dust, and it is not nothing. There is exactly one clean, provable speedup of genuine relevance — a quadratic (not exponential) acceleration of Monte Carlo via Quantum Amplitude Estimation — and a handful of optimization and machine-learning angles that range from “plausibly useful someday” to “almost certainly a dead end.” Your job by the end is to tell those apart cold, the way you learned to tell a real backtest from an overfit one in the ML courses. We assume you’ve internalized Monte Carlo for Finance (you’ll feel the QAE speedup in your bones only if you remember why classical MC error shrinks like $1/\sqrt{N}$) and Portfolio Optimization (QUBO and QAOA are just your mean–variance problem wearing a binary-variable costume). The arc:

• Quantum foundations for finance — qubits, superposition, entanglement and gates explained for the quant, not the physicist; and the two hardware paradigms you’ll keep meeting: the gate model (universal, runs algorithms like QAE/QAOA) versus quantum annealing (special-purpose, solves one kind of optimization problem).
• Amplitude estimation & Monte Carlo — the crown jewel: QAE turns classical MC’s $1/\sqrt{N}$ error decay into $1/N$, a genuine quadratic speedup for option pricing and risk (VaR/CVaR). The cleanest theoretical win in the whole field — with a giant asterisk we spend the rest of the course earning.
• QUBO & portfolio optimization — recasting asset selection as Quadratic Unconstrained Binary Optimization, solved by QAOA on the gate model or quantum annealing on D-Wave. Where this shines: the discrete, cardinality-constrained problems (pick exactly $k$ names, lot-size constraints) that make classical convex solvers choke.
• Quantum machine learning for signals — quantum kernels and variational classifiers for alpha, taught with the heaviest skepticism in the course: the theory is elegant, the finance evidence is thin, and the data-loading wall is brutal.
• The reality check: NISQ & data loading — the walls. Today’s noisy intermediate-scale quantum machines decohere before deep circuits finish; error correction demands thousands of physical qubits per logical one; and above all the data-loading / state-preparation problem can quietly erase your quadratic speedup before the algorithm even starts.
• Quantum-inspired classical methods — the punchline that pays rent today: tensor networks and other quantum-inspired algorithms that run on the classical hardware in your data center right now, and the discipline of judging any quantum claim end-to-end — loading and readout included — against the best classical baseline.

By the end you’ll be able to read a “quantum advantage in finance” claim and instantly locate it on the map: which speedup it invokes, whether the data-loading cost was honestly accounted for, and whether the end-to-end pipeline could ever beat a well-tuned classical method. A graded final exam runs the whole sober discipline back at you, one locked question at a time.