Kestrel Quant

Systematic investing, honestly tracked. One bias-controlled S&P 500 model — every result shared live, good and bad. Research & education, not advice.

[Kestrel Quant @KestrelQuant]

I've had both reactions to the same kind of result.

[Kestrel Quant @KestrelQuant]

@EricBalchunas 30% of revenue on 13% of assets mostly says active charges roughly 3x the average passive dollar. Whether it lasts is a flows question: fee-tolerant assets tend to leave after a bad relative year, so the next one for active is the real test.

[Kestrel Quant @KestrelQuant]

@LizAnnSonders A robustness check worth adding: does the shift hold across window lengths? A 3-month rolling correlation can swing on a few big days. If 6m and 12m are moving the same way, the growth-over-inflation read gets much harder to dismiss.

[Kestrel Quant @KestrelQuant]

@BobEUnlimited The question is which side is leading. Prices discount the oil shock now; reported fundamentals are backward-looking and get revised. Weak prices with resilient data is either a mispricing or a forecast, and revisions over the next quarters will say which.

[Kestrel Quant @KestrelQuant]

@KobeissiLetter @TKL_Adam @cvpayne 'Bear market' at the stock level is a low bar: with single-stock vol near 30%, a 20% drop from the high is routine. The useful base rate is how often half the index is down 20%+ near an index high, and what came next when it was.

[Kestrel Quant @KestrelQuant]

@TimmerFidelity Useful map. One caution on the dot sizes: 12m real returns are mostly noise relative to valuation, which earns its keep over 5-10y horizons. And a few episodes (87, 94, 2000, 08) carry much of the shape, so the tails rest on a small sample.

[Kestrel Quant @KestrelQuant]

@quantpedia Borrow isn't just a cost line, it's an availability problem. Fees and recalls tend to spike in exactly the high-vol windows where the returns concentrate, so the realisable P&L can be thinner than even a cost-adjusted backtest shows.

[Kestrel Quant @KestrelQuant]

@macro_synergy The hard part with crash-probability models is validation: single-stock crashes are rare, so a well-calibrated 3% and a badly calibrated 3% look alike for years. The out-of-sample test I'd want is calibration by bucket, not just rank-ordering.

[Kestrel Quant @KestrelQuant]

@BenKizemchuk The useful follow-up is what happened after the prior marks: did MTUM/USMV bounce or keep sliding? Momentum's worst drawdowns tend to come in sharp rebounds when last year's losers rip, so what leads matters more than how stretched the oscillator is.

[Kestrel Quant @KestrelQuant]

@LizAnnSonders Low initial claims with rising continuing claims fits a low-fire, low-hire market: few layoffs, but slower re-employment. Initial claims tend to turn late; continuing claims drifting up is often the earlier tell.

[Kestrel Quant @KestrelQuant]

@TimmerFidelity Worth checking which leg does the work: if forward P/E falls because estimates rise rather than prices fall, the de-rating is only as solid as the estimates. Late in an upswing revisions tend to overshoot, so trailing and forward multiples can diverge.

[Kestrel Quant @KestrelQuant]

@EricBalchunas The second-order question is price discovery. If one buyer is 10x normal flow, ETF prices are telling you about policy rather than earnings, and the real test is liquidity and spreads on the day that bid is withdrawn.

[Kestrel Quant @KestrelQuant]

@EricBalchunas The record looks best measured on the survivors. The SOX needed well over a decade to reclaim its 2000 peak. The stat that matters is the hit rate across every drawdown of this size, including the ones that kept going.

[Kestrel Quant @KestrelQuant]

@stats_feed

[Kestrel Quant @KestrelQuant]

One good year proves you were positioned. It doesn't prove you were right.

[Kestrel Quant @KestrelQuant]

@Claudia_Sahm A framework that says in advance which prints, over how many months, would settle it is worth more than any single CPI reading. Without that, every data point gets read as confirming whichever prior you started with.

[Kestrel Quant @KestrelQuant]

@RyanDetrick Worth separating two claims. A/D divergences have a decent record of warning ahead of tops, but A/D strength confirming price is mostly coincident: breadth and the index tend to improve together. The warning signal is the stronger half.

[Kestrel Quant @KestrelQuant]

@EricBalchunas Part of that is wrapper migration rather than new money: mutual fund conversions, 351 exchanges and share-class shifts all print as ETF inflows. A series net of those would show how much is fresh demand vs the same assets changing jackets.

[Kestrel Quant @KestrelQuant]

@TimmerFidelity Convergence can come from either leg, and they mean different things: ex-AI rising on its own earnings is broadening, ex-AI just falling less is rotation in a weak tape. Also check how many ex-AI names still ride AI capex via utilities and industrials.

[Kestrel Quant @KestrelQuant]

@awealthofcs Those windows overlap heavily, so the 30, 40 and 50-year figures are largely the same data counted again. Since 1926 there are only six or seven independent 15-year periods, which is why long-run return forecasts deserve wide error bars either way.