openquant.structural_breaks

Compiled from Rust (crates/pyopenquant). Signatures and types are from the shipped .pyi stubs; descriptions are the functions’ own docstrings, as help() shows them. All Python modules.

Functions

get_chow_type_stat

get_chow_type_stat(log_prices: Sequence[float], min_length: int) -> list[float]

Chow-type Dickey-Fuller statistics for a single break at each candidate date.

AFML §17.3.1. For each candidate break index tau from min_length to len(log_prices) - min_length - 1, regresses the log-price change dy_t on the lagged level y_{t-1} with the regressor zeroed before tau (no intercept), and records the t-statistic beta / se(beta). Large values suggest a switch from a random walk to an explosive process at tau. A singular regression gives NaN.

Parameters

log_prices : list[float]

Log prices, oldest first.

min_length : int

Minimum number of observations kept on each side of a candidate break.

Returns

list[float]

One statistic per candidate break, len(log_prices) - 2 * min_length values; empty if the series is shorter than 2 * min_length.

Raises

ValueError

If the regression has no rows (min_length == 0 with a single price).

get_chu_stinchcombe_white_statistics

get_chu_stinchcombe_white_statistics(
    log_prices: Sequence[float],
    test_type: str,
) -> tuple[list[float], list[float]]

Chu-Stinchcombe-White CUSUM test on levels, with its critical values.

AFML §17.3.2. For each bar t from index 2 on, sigma_t^2 is the mean of the squared one-bar changes up to t, and the statistic is S_t = max_{n < t} (y_t - y_n) / (sigma_t * sqrt(t - n)) (absolute difference for "two_sided"). The critical value is sqrt(4.6 + ln(t - n)) at the maximising n.

Parameters

log_prices : list[float]

Log prices, oldest first; at least three.

test_type : str

"one_sided" (tests for an upward departure) or "two_sided".

Returns

tuple[list[float], list[float]]

(critical_value, stat), each with len(log_prices) - 2 values, for bars 2 onward. Note the critical values come first.

Raises

ValueError

If there are fewer than three prices or test_type is not "one_sided" or "two_sided".

get_sadf

get_sadf(
    series: Sequence[float],
    model: str,
    add_const: bool,
    min_length: int,
    lags: int,
) -> list[float]

Supremum ADF and sub/super-martingale explosiveness statistics.

AFML §17.4.2-17.4.3, Snippets 17.1-17.4. Builds one regression row per bar from bar lags + 1 on, then returns, for each row from position min_length on, the supremum over every backward-expanding window of at least min_length rows ending at that row of the statistic below (t is the 0-based row position over the whole sample):

• "linear": dy on y_{t-1}, lags lagged dy, a constant (if add_const) and t; the t-statistic of y_{t-1}.
• "quadratic": as "linear" plus t^2 (Snippet 17.2 ctt).
• "sm_poly_1": y on 1, t, t^2; |beta(t^2)| / se.
• "sm_poly_2": log y on 1, t, t^2; |beta(t^2)| / se.
• "sm_exp": log y on 1, t; |beta(t)| / se.
• "sm_power": log y on 1, log(t + 1); |beta| / se.

The sm_* models ignore add_const and lags beyond setting the first row, take the absolute value because a trend of either sign is of interest, and need a positive series when they take logs. Windows whose regression is singular are skipped; a row with no usable window is -inf. Every window of every row is a separate regression, so the cost grows quickly with the series length.

Parameters

series : list[float]

Log prices (or prices, for sm_poly_1), oldest first.

model : str

One of "linear", "quadratic", "sm_poly_1", "sm_poly_2", "sm_exp", "sm_power".

add_const : bool

Whether the ADF models include a constant.

min_length : int

Minimum window length in regression rows; must be >= 1.

lags : int

Number of lagged differences in the ADF regression (lags 1..=lags).

Returns

list[float]

One statistic per regression row from position min_length on, i.e. len(series) - lags - 1 - min_length values; empty if there are no more than min_length rows.

Raises

ValueError

If model is unknown, the series has fewer than two values or no more than lags + 1, or min_length is 0 (an empty window).