File size: 15,275 Bytes
ebcde1f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
"""Reproducible descriptive economics for the frozen live-L2 study.

These functions run only after endpoint labels are opened.  They never select a
model or refit a threshold used by prediction.  Shock thresholds and stability
bins are fitted on the declared development reference, then applied unchanged
to held-out sessions.
"""

from __future__ import annotations

import math
from collections.abc import Sequence
from dataclasses import dataclass

import polars as pl

from microstructure.research.analysis import feature_stability_summary
from microstructure.research.l2_multidate import L2ResearchError, validate_l2_endpoint_frame

_NANOSECONDS_PER_MINUTE = 60_000_000_000
_DEVELOPMENT_ROLES = frozenset({"train", "validation"})
_TEST_ROLES = ("primary_test", "replication_test")


@dataclass(frozen=True, slots=True)
class L2DescriptiveAnalysis:
    """Machine-readable L2 economic analyses, kept separate from model scores."""

    intraday_liquidity: pl.DataFrame
    ofi_return_association: pl.DataFrame
    signal_half_life: pl.DataFrame
    liquidity_recovery: pl.DataFrame
    regime_diagnostics: pl.DataFrame
    feature_stability: pl.DataFrame
    cross_instrument_stability: pl.DataFrame


def _combined(frames: Sequence[pl.DataFrame]) -> pl.DataFrame:
    values = tuple(frames)
    if not values:
        raise L2ResearchError("L2 descriptive analysis requires endpoint frames")
    for frame in values:
        validate_l2_endpoint_frame(frame)
    combined = pl.concat(values, how="diagonal_relaxed")
    keys = ["study_date", "symbol", "endpoint_name", "sample_id"]
    if combined.select(keys).unique().height != combined.height:
        raise L2ResearchError("L2 descriptive endpoint identities are not unique")
    return combined


def _canonical_books(combined: pl.DataFrame) -> pl.DataFrame:
    endpoint_names = sorted(str(value) for value in combined.get_column("endpoint_name").unique())
    chosen = "event_20" if "event_20" in endpoint_names else endpoint_names[0]
    books = combined.filter(pl.col("endpoint_name") == chosen)
    keys = ["study_date", "symbol", "continuity_id", "decision_sequence"]
    if books.select(keys).unique().height != books.height:
        raise L2ResearchError("canonical L2 book observations are not unique")
    return books


def _intraday_liquidity(books: pl.DataFrame) -> pl.DataFrame:
    return (
        books.with_columns(
            ((pl.col("decision_ts_ns") // _NANOSECONDS_PER_MINUTE) % (24 * 60))
            .cast(pl.Int32)
            .alias("utc_minute_of_day")
        )
        .group_by("study_date", "study_role", "symbol", "utc_minute_of_day")
        .agg(
            pl.len().alias("n_observations"),
            pl.col("spread_bps").mean().alias("mean_spread_bps"),
            pl.col("spread_bps").median().alias("median_spread_bps"),
            pl.col("depth_total_l1").mean().alias("mean_depth_l1"),
            pl.col("depth_total_l5").mean().alias("mean_depth_l5"),
            pl.col("depth_total_l10").mean().alias("mean_depth_l10"),
            pl.col("queue_imbalance_l1").mean().alias("mean_queue_imbalance_l1"),
            pl.col("realized_volatility_w100").mean().alias("mean_realized_volatility_w100"),
        )
        .sort("study_date", "symbol", "utc_minute_of_day")
    )


def _finite_correlation(x: pl.Series, y: pl.Series) -> float | None:
    pairs = (
        pl.DataFrame({"x": x, "y": y})
        .drop_nulls()
        .filter(pl.col("x").is_finite() & pl.col("y").is_finite())
    )
    if (
        pairs.height < 3
        or pairs.get_column("x").n_unique() < 2
        or pairs.get_column("y").n_unique() < 2
    ):
        return None
    value = pairs.select(pl.corr("x", "y")).item()
    return float(value) if value is not None and math.isfinite(float(value)) else None


def _ofi_association(combined: pl.DataFrame) -> pl.DataFrame:
    rows: list[dict[str, object]] = []
    for key, frame in combined.filter(~pl.col("right_censored")).group_by(
        "study_date", "study_role", "symbol", "endpoint_name", maintain_order=True
    ):
        study_date, study_role, symbol, endpoint_name = (str(value) for value in key)
        side_source = str(frame.get_column("signed_markout_side_source")[0])
        rows.append(
            {
                "study_date": study_date,
                "study_role": study_role,
                "symbol": symbol,
                "endpoint_name": endpoint_name,
                "side_source": side_source,
                "n_observations": frame.height,
                "ofi_return_correlation": _finite_correlation(
                    frame.get_column(side_source), frame.get_column("future_mid_return")
                ),
                "mean_ofi_signed_future_mid_markout_bps": frame.get_column(
                    "ofi_signed_future_mid_markout_bps"
                ).mean(),
                "positive_direction_rate": frame.get_column("future_mid_up").mean(),
                "interpretation": "descriptive_book_flow_markout_not_trade_impact",
            }
        )
    return pl.DataFrame(rows, infer_schema_length=None).sort(
        "study_date", "symbol", "endpoint_name"
    )


def _signal_half_life(association: pl.DataFrame, combined: pl.DataFrame) -> pl.DataFrame:
    endpoint_contract = (
        combined.select(
            "endpoint_name", "endpoint_domain", "endpoint_horizon_value", "endpoint_horizon_unit"
        )
        .unique()
        .sort("endpoint_domain", "endpoint_horizon_value")
    )
    enriched = association.join(endpoint_contract, on="endpoint_name", how="left")
    rows: list[dict[str, object]] = []
    for key, frame in enriched.group_by(
        "study_date", "study_role", "symbol", "endpoint_domain", maintain_order=True
    ):
        study_date, study_role, symbol, domain = (str(value) for value in key)
        ordered = frame.sort("endpoint_horizon_value")
        correlations = ordered.get_column("ofi_return_correlation").to_list()
        baseline = (
            abs(float(correlations[0])) if correlations and correlations[0] is not None else None
        )
        crossed: int | None = None
        if baseline is not None and baseline > 0:
            for horizon, correlation in zip(
                ordered.get_column("endpoint_horizon_value").to_list(),
                correlations,
                strict=True,
            ):
                if correlation is not None and abs(float(correlation)) <= 0.5 * baseline:
                    crossed = int(horizon)
                    break
        for row in ordered.to_dicts():
            correlation = row["ofi_return_correlation"]
            association_ratio = None
            if correlation is not None and baseline is not None and baseline != 0.0:
                association_ratio = abs(float(correlation)) / baseline
            rows.append(
                {
                    "study_date": study_date,
                    "study_role": study_role,
                    "symbol": symbol,
                    "endpoint_domain": domain,
                    "endpoint_name": row["endpoint_name"],
                    "horizon_value": row["endpoint_horizon_value"],
                    "horizon_unit": row["endpoint_horizon_unit"],
                    "ofi_return_correlation": correlation,
                    "absolute_association_ratio_to_shortest": association_ratio,
                    "half_life_crossed_at_horizon": crossed,
                    "half_life_status": (
                        "crossed_within_declared_horizons"
                        if crossed is not None
                        else "not_crossed_or_unidentified"
                    ),
                }
            )
    return pl.DataFrame(rows, infer_schema_length=None).sort(
        "study_date", "symbol", "endpoint_domain", "horizon_value"
    )


def _training_shock_thresholds(books: pl.DataFrame) -> pl.DataFrame:
    train = books.filter(pl.col("study_role") == "train").with_columns(
        pl.min_horizontal("bid_quantity", "ask_quantity").alias("executable_l1_depth")
    )
    symbols = set(str(value) for value in books.get_column("symbol").unique())
    if set(str(value) for value in train.get_column("symbol").unique()) != symbols:
        raise L2ResearchError("liquidity shock thresholds require train rows for every symbol")
    return train.group_by("symbol").agg(
        pl.col("spread_bps").quantile(0.95, interpolation="linear").alias("train_spread_q95"),
        pl.col("executable_l1_depth")
        .quantile(0.05, interpolation="linear")
        .alias("train_executable_depth_q05"),
    )


def _liquidity_recovery(books: pl.DataFrame) -> pl.DataFrame:
    thresholds = _training_shock_thresholds(books)
    joined = books.with_columns(
        pl.min_horizontal("bid_quantity", "ask_quantity").alias("executable_l1_depth")
    ).join(thresholds, on="symbol", how="left")
    group = ["study_date", "symbol", "continuity_id"]
    rows: list[pl.DataFrame] = []
    for horizon in (20, 100):
        rows.append(
            joined.with_columns(
                pl.col("spread_bps").shift(-horizon).over(group).alias("future_spread_bps"),
                pl.col("executable_l1_depth")
                .shift(-horizon)
                .over(group)
                .alias("future_executable_l1_depth"),
            )
            .filter(
                (pl.col("spread_bps") >= pl.col("train_spread_q95"))
                | (pl.col("executable_l1_depth") <= pl.col("train_executable_depth_q05"))
            )
            .drop_nulls(["future_spread_bps", "future_executable_l1_depth"])
            .with_columns(
                pl.lit(horizon).alias("recovery_horizon_events"),
                (pl.col("future_spread_bps") - pl.col("spread_bps")).alias("spread_change_bps"),
                (pl.col("future_executable_l1_depth") - pl.col("executable_l1_depth")).alias(
                    "executable_depth_change"
                ),
            )
            .group_by("study_date", "study_role", "symbol", "recovery_horizon_events")
            .agg(
                pl.len().alias("n_shocks"),
                pl.col("spread_change_bps").mean().alias("mean_spread_change_bps"),
                pl.col("executable_depth_change").mean().alias("mean_executable_depth_change"),
                pl.col("train_spread_q95").first(),
                pl.col("train_executable_depth_q05").first(),
            )
        )
    return pl.concat(rows, how="vertical_relaxed").sort(
        "study_date", "symbol", "recovery_horizon_events"
    )


def _regime_diagnostics(combined: pl.DataFrame) -> pl.DataFrame:
    required = {"volatility_regime", "liquidity_regime"}
    missing = sorted(required.difference(combined.columns))
    if missing:
        raise L2ResearchError(f"regime diagnostics are missing columns: {missing}")
    return (
        combined.filter(~pl.col("right_censored"))
        .group_by(
            "study_date",
            "study_role",
            "symbol",
            "endpoint_name",
            "volatility_regime",
            "liquidity_regime",
        )
        .agg(
            pl.len().alias("n_observations"),
            pl.col("future_mid_up").mean().alias("positive_direction_rate"),
            pl.col("future_mid_return").mean().alias("mean_future_mid_return"),
            pl.col("ofi_signed_future_mid_markout_bps")
            .mean()
            .alias("mean_ofi_signed_future_mid_markout_bps"),
        )
        .sort("study_date", "symbol", "endpoint_name", "volatility_regime", "liquidity_regime")
    )


def _feature_stability(
    combined: pl.DataFrame, *, feature_columns: tuple[str, ...], bins: int
) -> pl.DataFrame:
    reference = combined.filter(pl.col("study_role").is_in(list(_DEVELOPMENT_ROLES)))
    if reference.is_empty():
        raise L2ResearchError("feature stability requires development rows")
    outputs: list[pl.DataFrame] = []
    for role in _TEST_ROLES:
        comparison = combined.filter(pl.col("study_role") == role)
        if comparison.is_empty():
            raise L2ResearchError(f"feature stability requires {role} rows")
        outputs.append(
            feature_stability_summary(
                reference,
                comparison,
                feature_columns=feature_columns,
                group_columns=("symbol", "endpoint_name"),
                bins=bins,
            ).with_columns(
                pl.lit(role).alias("comparison_role"),
                pl.lit("train_plus_validation_only").alias("reference_scope"),
            )
        )
    return pl.concat(outputs, how="vertical_relaxed").sort(
        "comparison_role", "symbol", "endpoint_name", "feature"
    )


def _cross_instrument_stability(association: pl.DataFrame) -> pl.DataFrame:
    rows: list[dict[str, object]] = []
    for key, frame in association.group_by(
        "study_date", "study_role", "endpoint_name", maintain_order=True
    ):
        study_date, study_role, endpoint_name = (str(value) for value in key)
        by_symbol: dict[str, float | None] = {
            str(row["symbol"]): (
                None
                if row["ofi_return_correlation"] is None
                else float(row["ofi_return_correlation"])
            )
            for row in frame.to_dicts()
        }
        btc = by_symbol.get("BTCUSDT")
        eth = by_symbol.get("ETHUSDT")
        same_direction = None
        if btc is not None and eth is not None and btc != 0.0 and eth != 0.0:
            same_direction = math.copysign(1.0, btc) == math.copysign(1.0, eth)
        rows.append(
            {
                "study_date": study_date,
                "study_role": study_role,
                "endpoint_name": endpoint_name,
                "btc_ofi_return_correlation": btc,
                "eth_ofi_return_correlation": eth,
                "both_observed": btc is not None and eth is not None,
                "same_direction": same_direction,
                "cross_instrument_pooling": False,
            }
        )
    return pl.DataFrame(rows, infer_schema_length=None).sort("study_date", "endpoint_name")


def build_l2_descriptive_analysis(
    endpoint_frames: Sequence[pl.DataFrame],
    *,
    feature_columns: Sequence[str],
    stability_bins: int,
) -> L2DescriptiveAnalysis:
    """Build all frozen descriptive outputs without fitting a predictive model."""

    features = tuple(str(value) for value in feature_columns)
    if not features or stability_bins < 2:
        raise L2ResearchError("descriptive feature columns and stability bins are required")
    combined = _combined(endpoint_frames)
    books = _canonical_books(combined)
    association = _ofi_association(combined)
    return L2DescriptiveAnalysis(
        intraday_liquidity=_intraday_liquidity(books),
        ofi_return_association=association,
        signal_half_life=_signal_half_life(association, combined),
        liquidity_recovery=_liquidity_recovery(books),
        regime_diagnostics=_regime_diagnostics(combined),
        feature_stability=_feature_stability(
            combined, feature_columns=features, bins=stability_bins
        ),
        cross_instrument_stability=_cross_instrument_stability(association),
    )


__all__ = ["L2DescriptiveAnalysis", "build_l2_descriptive_analysis"]