quantitative-finance

If your cost per round trip equals the move you are trying to capture, you need 100 percent directional accuracy to break even. This lesson computes that hurdle, replaces accuracy with metrics tied to a tradeable decision, and covers the capacity and latency limits that decide whether a real edge is worth anything.

At a one-event horizon, 92 percent of mid-price labels are exactly no-change, so a model that always predicts flat scores 92 percent accuracy. This lesson computes that baseline across horizons and works through the four mechanisms that turn a genuine measurement into a number nobody can reproduce.

Convolutional and recurrent networks have been applied to order book data with published success, and the architectures encode real assumptions about the book's structure. This lesson explains what each one assumes, why the gains over simple baselines are smaller than headline numbers suggest, and where the modelling effort is better spent.

Before choosing an architecture you have to decide what the input is, and an order book offers several incompatible representations that are not equally informative. This lesson covers what each data level contains, why raw prices are the wrong features, and the representation choices that decide more than the model does.

A market maker quotes both sides and profits from the spread, but every fill leaves an unwanted position and the counterparties who trade most eagerly are the ones who know something. This lesson simulates the inventory-skew trade-off, showing a 63 percent cut in exposure for 2.5 percent of profit, and locates where a learned policy genuinely helps.

A static execution schedule is an open-loop policy: it commits to a plan before seeing anything. This lesson formulates execution as an MDP so the plan can react, and computes the ceiling on what any adaptive policy could win, which turns out to collapse as market impact grows.

Shuffled cross-validation reports 54 percent accuracy on data containing no signal, because neighbouring samples share their futures. This lesson builds purging, embargo and sample uniqueness weights, shows the illusion scaling with label overlap and vanishing when the fix is applied, and covers why a single walk-forward path is weak evidence.

Prices are non-stationary and returns are stationary but forget everything, so the standard fix throws away the level information a model needed. This lesson builds fractional differentiation, which makes the choice a dial rather than a switch, and shows by computation that most of the memory can be kept while still passing a stationarity test.

The default label, the sign of the return over a fixed horizon, describes a trade nobody would take: no stop, no target, and a holding period chosen by the modeller rather than by the market. This lesson builds the triple-barrier method and meta-labelling, and shows with computed numbers how much of a fixed-horizon dataset is fiction.

Machine learning practice assumes independent samples, a stable data-generating process, and an environment that does not react to you. Financial data violates all three, and the result is models that validate beautifully and lose money. This lesson establishes what is actually broken before the rest of the course fixes it.

The most prominent study in this area found a language model predicting stock reactions from headlines, and it is usually reported as proving something it explicitly does not claim. This lesson reads it precisely, follows its qualifiers to their consequences, and shows why a real statistical result and a tradable strategy are different things.

Backtesting a language model signal has a defect no other signal has. The model was trained on text from the period being tested, so it may already know what happened next, and it uses that knowledge even when instructed not to. This lesson establishes the problem from the published evidence, shows why prompting does not fix it, and covers what does.

The popular framing is a model that predicts prices. That is the one job the technology is worst suited to, and it obscures the one it is genuinely good at: turning unstructured text into structured data at a scale that was previously unaffordable. This lesson locates LLMs against the trading stack and rules out the places they cannot go.

The edge that survives statistics still has to survive trading. Spread, market impact and the square-root law, why every strategy has a capital ceiling, and the research protocol that makes a backtest worth believing.

The statistical core of backtest overfitting: why the best of many trials is inflated even when nothing works, how much to discount it, and why finance needs a far higher significance bar than the usual one.

Look-ahead bias, survivorship bias, and point-in-time data. The errors that make a backtest wrong as a simulation, independent of any statistical question about whether the edge is real.

A backtest is not a measurement of the past, it is a counterfactual about a world that never happened. Getting precise about that claim explains every way backtests mislead.