04 · TICKER
COST992.13— 1y
Nothing stored for COST
COST is on a covered watchlist, so the next run will fetch it — but no run has stored bars for it yet. Pulling it adds COST to the watchlist and dispatches a backfill for that one name.
runs on the compute plane · minutes, not seconds · writes must be unlocked
SPOT
992.13
2026-10-27 close
1Y RETURN
—
not enough history
YZ VOL 22D
18.1%
annualized · 2026-10-27
VOL PCTL 1Y
17th
stored by the nightly · as of 2026-10-27
BETA · 1Y
-0.18
vs SPY daily
CORR · 1Y
-0.12
vs SPY daily
DRIFT ν · ANNUAL
—
0 measured daily returns — a drift estimate needs at least 252, and even that carries a wide interval
MAP REGIME
—
no fit
PRICE · DAILY CLOSE · VOLUME AT PRICE
last —
no stored bars
SUPPORT & RESISTANCE · COST
no stored bars to profile
REGIME · COST
—
no regime fit stored
■ Bear market · BEAR · no days measured■ Bear rally · RALLY · no days measured■ Bull correction · CORR · no days measured■ Bull market · BULL · no days measured
The percentage beside each name is what that state’s own days did — the annualized mean log total return of the sessions this fit assigns to it. The four names are the Maheu–McCurdy labels and are not re-derived from the data, so on a given ticker a state can carry a return its name does not suggest; that number is how a reader sees it without having to work it out. It is a measurement of the past, not a property of the state. no per-ticker fit stored — the market-level fit on 03 is not substituted here
REALIZED VOL · YANG–ZHANG
— 11d — — 22d — — 44d —
no volatility series stored
CONE · COST · 60 SESSIONS
a cone needs spot, a volatility estimate and a drift estimate — one of them is not measured
VOLATILITY MODELS · COST · 22D WINDOW, ANNUALIZED
ESTIMATOR
VOL
BASIS
Close-to-close
—
std of log close-to-close returns · ignores intraday range
Parkinson
—
high–low range · assumes no drift, no jumps
Garman–Klass
—
OHLC · more efficient, biased by overnight gaps
Rogers–Satchell
—
drift-independent OHLC · no overnight component
Yang–Zhang
—
overnight + k·close + (1−k)·Rogers–Satchell
GARCH(1,1)
—
MLE fit · conditional forecast at the stated horizon
Yang–Zhang is the house estimator; the others are shown because their disagreement is itself a measurement — of gap risk, drift and jumps — rather than noise to be averaged away.
DISTRIBUTION FIT · DAILY LOG RETURNS
STUDENT-T DF
—
KS p-VALUE
—
EFFECTIVE N
—
GRADE
—
No distribution fit is stored for this ticker, so the Student-t column below is absent rather than defaulted.
fit as of — · KS at α .05 against the fitted Student-t, computed against the effective sample size (— independent-equivalent from — overlapping windows) — testing on the raw count rejects normality about five times too often
POSITIONS IN COST · 0 OPEN
no book imported on this device — 01 has the import flow
DRIFT
estimate — ± — s.e.
regime tilt —
ν used —
long window · 1260 sessions
WHAT ν IS
ν is the annualized drift of the log price: the exponential growth rate that positions the median path of the cone above, and the centre every terminal and touch probability on this page is measured from. It is not the expected simple return — that is ν + ½σ², — here — and nothing in this system adds the ½σ² twice.
HOW THIS ONE WAS DETERMINED
The mean daily log return over the last 0 stored sessions, multiplied by 252, plus a regime tilt. The tilt is the posterior-weighted deviation of the fitted state means from their occupancy-weighted unconditional mean — derived from the stored fit, never a constant table — and it is absent rather than zero when the fit recorded no state means. Nothing shorter is offered: annualizing a 5- or 20-day mean multiplies its standard error by 252, and the audit measured that ensemble’s noise at sixteen times the signal.
FIT QUALITY OF THE TRAILING MEAN
0 measured daily returns — a drift estimate needs at least 252, and even that carries a wide interval
t is the trailing mean over its own standard error, and p is its two-sided normal tail — daily log returns are close enough to serially uncorrelated for a mean’s standard error, which is not true of the overlapping windows the study statistics use. Years-to-detect is (1.96σ/ν)², the sample length at which an estimate this size would clear 95%. The regime tilt is posterior-weighted and pivots on the occupancy-weighted unconditional mean.
TARGET-PRICE PROBABILITY · COST
empirical · student-t · lognormal — computed here, from the controls
SPOT
992.13
TARGET
HORIZON45 days
VOLATILITY
using — ann · Yang–Zhang 22d · as of 2026-10-27
TERMINAL · P(FINISHES BEYOND THE TARGET)
what prices a payoff
EMPIRICAL · BOOTSTRAP 4,000 PATHS
—
0 measured daily returns — below the 250 the bootstrap needs
STUDENT-T · DF —
—
no stored Student-t fit for this ticker
LOGNORMAL
—
spot, volatility or drift is not measured
TOUCH · P(TRADES THROUGH THE TARGET AT ANY POINT)
what a strike or a stop gets tested by
EMPIRICAL · BOOTSTRAP 4,000 PATHS
—
0 measured daily returns — below the 250 the bootstrap needs
STUDENT-T · DF —
—
no first-passage law exists for a scaled Student-t increment — the empirical path simulation is the touch estimate here
LOGNORMAL
—
spot, volatility or drift is not measured
COST vs 1111.19 IN 45D
These are two different questions, and they differ by roughly a factor of two. Terminal is the probability of finishing beyond the target at the horizon — the number that prices a payoff. Touch is the probability of trading through it at any point before then — the number a strike or a stop is actually exposed to. Within each group the three rows share one drift and one volatility, so the spread between them is a statement about distributional shape and nothing else.
expected move ±1σ: — · log-return needed: +11.33% · ν — · σ —
DATA QUALITY · COST
measuring the stored bars…