Factor Analysis Report

Diagnostic factor tear sheet.

Methodology & Return Provenance

Factor forward returns are diagnostic labels and must not be interpreted as realized fill-to-exit strategy returns.

Report calculation contract

Internal research report

Backtest engine

research backtester

Factor formation schedule

Month start (first trading session); market open +60 minutes

Observed factor-formation cadence

approximately monthly; median 31.0 calendar days across 32 report formations.

Usable IC cadence (5D)

approximately monthly; median 31.0 calendar days across 32 usable IC observations.

Rolling IC window

Target up to 12 usable formation observations (~12 calendar months); partial rolling means begin at 5 observations. This is separate from the 5D forward-return horizon.

Stored signal timestamp

Formation session date only; the exact intraday callback timestamp is not retained in the factor-report dataset.

Forward-return entry

Formation-session close at the factor index date (price[t]).

Forward-return endpoint

Close N available daily price-matrix rows after formation (price[t+N]); a missing endpoint produces a missing label. N is not a calendar-day count.

Price field

close

Price basis

Adjusted close

Corporate-action adjustment

Research data portal corporate-action adjustments

Price source

research data portal

Intended-target return attribution

Next available return session after the target date for intended-target daily return attribution only; this is not an order or fill timestamp.

Execution / realized returns

Orders are submitted during the configured rebalance callback; actual transaction timestamps and fill prices are authoritative. Factor labels are not realized fill-to-exit returns.

Quantile-return benchmark

At each formation, subtract the equal-weight mean return of all eligible assets with a valid return for that horizon.

Quantile-return aggregation

Pool valid asset observations by quantile after formation-level demeaning; each asset observation has equal weight, so formations with more valid assets contribute more observations.

Top-minus-bottom spread aggregation

Within each formation, compute and horizon-normalize the top and bottom quantile means, subtract bottom from top, then average formation spreads with equal formation weight.

Horizon normalization

Convert each demeaned holding-horizon return to a 5D-equivalent compounded rate. These are relative rates, not absolute cumulative returns.

Information coefficient

Formation-date Spearman rank correlation with no group adjustment (group_adjust=False, by_group=False).

IC inference

Observation count is the number of usable formation dates. Mean-IC significance uses horizon-aware Newey-West/HAC inference rather than the legacy iid one-sample t-test.

Absolute Cumulative Forward Returns by Factor Quantile

Arithmetic close-to-close forward returns averaged within each quantile; these values are not cross-sectionally demeaned or horizon-normalized.

Cross-Sectionally Demeaned Equivalent Returns by Factor Quantile

Returns are demeaned against the equal-weight eligible universe at each formation, pooled as asset observations within quantile, and converted to 5D-equivalent compounded rates. They are relative rates, not absolute cumulative returns.

Information Coefficient Time Series (5D)

Each of the 32 usable IC observations is one formation-date Spearman rank correlation for the 5D forward-return horizon. The rolling line targets up to 12 usable formation observations (~12 calendar months at the approximately monthly usable-IC cadence); partial rolling means begin at 5 observations. The observed factor-formation cadence is approximately monthly. 5D describes the forward-return horizon; it does not describe the formation cadence or rolling-window length.

Information Analysis

Each IC observation is one usable formation date. Mean-IC t-statistics and p-values use horizon-aware Newey-West/HAC inference, which accounts for overlap and can be more conservative than the legacy iid one-sample t-test. No group adjustment is applied (group_adjust=False, by_group=False).

Returns Analysis

Alpha and beta are regression diagnostics. Quantile rows are cross-sectionally demeaned 5D-equivalent rates in basis points.

Quantile Mean Turnover

Factor Rank Autocorrelation

Observation Counts