gmp/f - f - Gitea

gmp/f

Fork 0

Commit Graph

Author	SHA1	Message	Date
Greg Pomerantz	89674c24dd	Compute all alphas in excess of the 3-mo T-bill rate (BIL) Raw-intercept alphas absorbed the T-bill yield on uninvested/levered portions (582 well-fitted funds >2%/yr off; sum-of-betas polluted by level-matching). Now fund AND sleeves are netted against BIL daily total return before every regression; a cash position contributes exactly zero. - decompose: rf_series()/excess(); shv+bil dropped from regressors (~0 columns in excess space); FULL_WINDOW -> 2007-06-01 (BIL inception; mixing raw pre-2007 with excess breaks the fit). - factors: same excess treatment; shv out of DRIVERS. - CASH axis redefined: alpha/cash_yield -> net cash position = 1 - sum(betas) (label 'cash (net posn)'). - CANDIDATE list 250 -> 11: the old list was mostly under-invested funds whose 'alpha' was cash yield, not skill. - refback.py: per-fund fitted reference (forward-selected sleeves) stored as ref_5y/ref_full in search_all.json; app alpha-search table gains a 'reference (5y)' column - the answer to 'what is alpha computed against' (the fund's OWN fitted sleeve mix, not one index). - App captions updated; raw-alpha-era results backed up as *_rawalpha.json (not deleted).	2026-08-30 15:50:15 -04:00
Greg Pomerantz	db5fc4626d	Fund Lab: returns-based strategy decomposition for the 16 shortlist funds - fundlab/decompose.py: per-fund OLS forward selection (BIC-gated, \|t\|>2, per-model complete cases so differently-vintaged candidates stay comparable) against curated DISTINCT-AXIS candidate sets; full-history + last-5y models; rolling 1y beta drift for static-vs-time-varying verdicts; per-fund holdings cross-check notes - results (13 unique funds; pmfkx/lcrix/egrsx are share classes): * jlpsx ~1.04x S&P 500, R2 0.96 5y (cleanest) * lamhx S&P + value/mid tilt, R2 0.95, stable * cosix 5y: HY +0.30 / MBS +0.29 / IG +0.18, R2 0.86 * cvsix market neutral, 5y R2 0.74, +5.5%/yr alpha (t 6.7) * pmaix multi-asset: HY .62 / EFA .23 / comm .05 / bonds -.15 * mbxix hedge: ivv .39 / ief -.67 / fxe -.28, R2 0.53 * atesx NOT a static mix - rolling beta to its own QQQ/SPY holdings is 0.13-0.89 (median 0.30): the 'risk managed' overlay is real * qspnx/svarx/eagmx/atrfx/pmorx: market-neutral or idiosyncratic - alpha, not sleeves (qspnx +12.8%/yr alpha t 4.0) * lcorx/lcrix: new classes (Jul 2026), no history yet - holdings only - atesx holdings: pulled from the adviser's SOI PDF (anchor-soi-5.31.26): QQQ 65.2% + SPY 29.3% + MMF 0.6%, options overlay 4.9% - pool: added qqq (Nasdaq 100) - needed to fit tech-concentrated funds - app Fund Lab tab: per-fund decomposition (verdict, R2 full/5y, alpha, tracking error, beta drift, component table + bar chart, holdings cross-check note) and an all-funds summary expander - tests: ols/forward-select engine tests (50/50 fundlab)	2026-08-26 11:59:59 -04:00

Author

SHA1

Message

Date

Greg Pomerantz

89674c24dd

Compute all alphas in excess of the 3-mo T-bill rate (BIL)

Raw-intercept alphas absorbed the T-bill yield on uninvested/levered
portions (582 well-fitted funds >2%/yr off; sum-of-betas polluted by
level-matching). Now fund AND sleeves are netted against BIL daily
total return before every regression; a cash position contributes
exactly zero.

- decompose: rf_series()/excess(); shv+bil dropped from regressors
  (~0 columns in excess space); FULL_WINDOW -> 2007-06-01 (BIL
  inception; mixing raw pre-2007 with excess breaks the fit).
- factors: same excess treatment; shv out of DRIVERS.
- CASH axis redefined: alpha/cash_yield -> net cash position = 1 -
  sum(betas) (label 'cash (net posn)').
- CANDIDATE list 250 -> 11: the old list was mostly under-invested
  funds whose 'alpha' was cash yield, not skill.
- refback.py: per-fund fitted reference (forward-selected sleeves)
  stored as ref_5y/ref_full in search_all.json; app alpha-search
  table gains a 'reference (5y)' column - the answer to 'what is
  alpha computed against' (the fund's OWN fitted sleeve mix, not one
  index).
- App captions updated; raw-alpha-era results backed up as
  *_rawalpha.json (not deleted).

2026-08-30 15:50:15 -04:00

Greg Pomerantz

db5fc4626d

Fund Lab: returns-based strategy decomposition for the 16 shortlist funds

- fundlab/decompose.py: per-fund OLS forward selection (BIC-gated, |t|>2,
  per-model complete cases so differently-vintaged candidates stay
  comparable) against curated DISTINCT-AXIS candidate sets; full-history
  + last-5y models; rolling 1y beta drift for static-vs-time-varying
  verdicts; per-fund holdings cross-check notes
- results (13 unique funds; pmfkx/lcrix/egrsx are share classes):
  * jlpsx  ~1.04x S&P 500, R2 0.96 5y  (cleanest)
  * lamhx  S&P + value/mid tilt, R2 0.95, stable
  * cosix  5y: HY +0.30 / MBS +0.29 / IG +0.18, R2 0.86
  * cvsix  market neutral, 5y R2 0.74, +5.5%/yr alpha (t 6.7)
  * pmaix  multi-asset: HY .62 / EFA .23 / comm .05 / bonds -.15
  * mbxix  hedge: ivv .39 / ief -.67 / fxe -.28, R2 0.53
  * atesx  NOT a static mix - rolling beta to its own QQQ/SPY holdings
            is 0.13-0.89 (median 0.30): the 'risk managed' overlay is real
  * qspnx/svarx/eagmx/atrfx/pmorx: market-neutral or idiosyncratic -
            alpha, not sleeves (qspnx +12.8%/yr alpha t 4.0)
  * lcorx/lcrix: new classes (Jul 2026), no history yet - holdings only
- atesx holdings: pulled from the adviser's SOI PDF (anchor-soi-5.31.26):
  QQQ 65.2% + SPY 29.3% + MMF 0.6%, options overlay 4.9%
- pool: added qqq (Nasdaq 100) - needed to fit tech-concentrated funds
- app Fund Lab tab: per-fund decomposition (verdict, R2 full/5y, alpha,
  tracking error, beta drift, component table + bar chart, holdings
  cross-check note) and an all-funds summary expander
- tests: ols/forward-select engine tests (50/50 fundlab)

2026-08-26 11:59:59 -04:00

2 Commits