| name | factor-exposure |
| description | Use when the user asks "is this alpha or just beta?", "what am I actually exposed to?", "factor analysis", "am I just riding the market?", or any variation about distinguishing skill from market exposure. Also use proactively when a group shows strong returns while the broad market is also up.
|
Factor Exposure
A group in Upward Trend with strong returns may just be riding the
broad market. Factor regression separates genuine alpha (the return
you earned beyond what the market explains) from beta (your exposure
to market movements). If alpha is zero after controlling for beta,
the Livermore signals are not adding edge in that group.
The script
scripts/factor.py regresses each leader's returns against SPY.
./run .claude/skills/factor-exposure/scripts/factor.py ~/.lafmm/data/semis
./run .claude/skills/factor-exposure/scripts/factor.py ~/.lafmm/data/energy --json
Options:
--benchmark PATH — ticker directory for the benchmark. Defaults to
data/us-indices/SPY (auto-discovered). Override for non-US markets
or sector-relative analysis.--json — machine-readable output.
Interpreting the output
Alpha (annualized): excess return not explained by market exposure.
Positive alpha means the group outperforms what its beta alone would
predict. Reported as a percentage per year.
Alpha t-statistic: statistical significance of the alpha estimate.
|t| > 1.96 is significant at 95% confidence (marked **).
|t| > 2.58 is significant at 99% confidence (marked ***).
If not significant, you cannot distinguish the alpha from zero.
Beta: sensitivity to market moves. Beta 1.0 means the stock moves
1:1 with the market. Beta 1.5 means it moves 50% more than the market
in both directions. A "high return" from a beta-2.0 stock in a rising
market is not alpha. It's leveraged beta.
R-squared: fraction of the stock's return variance explained by
the market. R² of 0.80 means 80% of the stock's moves are market-driven.
Only the remaining 20% is stock-specific.
Requirements
Both the group and the benchmark (SPY) need overlapping price history.
The regression needs at least 30 overlapping dates to produce
meaningful results. 60+ is preferable. The us-indices group provides
SPY data by default.
When to use this
- When a group is performing well and you want to know if it's genuine
edge or just market correlation.
- When comparing two groups with similar returns. The one with higher
alpha and lower beta has the real edge.
- Before sizing a position based on validate-signal results. A signal
with high Sharpe but pure beta exposure is not worth the same as one
with genuine alpha.
- When the user's portfolio seems concentrated. If all groups have
beta > 1.5, the portfolio is effectively leveraged to the market
regardless of how many "sectors" it holds.
Limitations
This is a single-factor model (market only). Adding sector ETFs as
additional factors would give a cleaner decomposition but requires
sector-specific benchmark data. The current implementation supports
additional factors via the library API (factor_regression(strategy, market, factors=[sector_returns])) but the CLI wrapper uses market
only.
Alpha is backward-looking. A stock that showed alpha over the last
90 days may not show it over the next 90. Use rolling_alpha from the
library to check whether alpha is stable or decaying.
