Module research 

Research universe and shared helpers for the backtesters.

Structs

Market

Market context from SPY, joined by date.

Rng

Small, fast, seedable PRNG (SplitMix64) with Gaussian draws.

Sample

One (symbol, bar) observation.

Scorecard

Directional scorecard for a set of (signed call strength, realized return).

Series

A loaded symbol: name plus dated bars.

Enums

Split

Which slice of history a sample belongs to.

Constants

BARRIER_ATR

Triple-barrier settings: which of ±BARRIER_ATR·ATR is touched first within BARRIER_BARS bars. This is the cleanest form of the question “which way does the squeeze break?”.

BARRIER_BARS

HORIZONS

Forecast horizons in trading days: one week, two weeks, one month.

MAGNITUDE_BASELINE

How many leading MAGNITUDE_NAMES form the volatility-only baseline.

MAGNITUDE_NAMES

Names of the expected-movement inputs, in magnitude_inputs order.

MARKET_FEATURES

UNIVERSE

A deliberately mixed research universe: index and sector ETFs, mega caps, cyclicals, defensives, laggards and a few high-volatility names. It is not the investable portfolio; it exists so the indicator is judged on many kinds of price behaviour, including stocks that went nowhere.

Functions

add_noise

Add Gaussian measurement noise to a price history.

auc

Area under the ROC curve: the chance a random “up” row scores higher than a random “down” row. 0.5 = no skill, regardless of base rate.

barrier_label

First-touch label. Uses the real future highs and lows.

build_samples

Build samples for one symbol. signal_bars is what the indicator sees (maybe noisy); truth is the real history used for outcomes.

load_universe

magnitude_inputs

Direction-free inputs for the expected-movement model.

pearson

scorecard

Grade calls. strength[i] > 0 is an up call, < 0 a down call, 0 no call. Only rows with |strength| >= threshold are graded.

spearman

Spearman rank correlation.

train_end

valid_end