SIGNAL H3 Health ↝ Home ↝ Harmony ↭ Wealth

Upcoming Top 50 has a check called GSADF Bubble Detection. Here's what it actually does, without the jargon. Ordinary statistics assume a stock's price wanders around randomly, like a random walk. GSADF asks a sharper question: has this price series recently started growing FASTER than a random walk would predict - the kind of explosive, self-reinforcing move that shows up right before (or during) a speculative run? It does this with a real statistical test (a recursive right-tailed unit-root regression), not a vibe check. The honest caveat: the exact critical value tables from the original research paper (Phillips-Shi-Yu) are calibrated by Monte Carlo simulation for each specific sample size - this app doesn't have those tables, so it uses a documented practical threshold instead. Treat the underlying t-stat as a ranking signal first, and the pass/fail flag as a convenience on top of it, not a certified statistical verdict.

Upcoming Top 50's Stealth Accumulation check is built on something called the Amihud illiquidity ratio - price impact per dollar of volume. The idea: if a huge amount of money is trading a stock and the price barely moves, something (or someone) is quietly absorbing all that volume without needing the price to move. That's very different from a stock that moves 5% on the same volume - which suggests thin, nervous liquidity, not steady accumulation. This app computes it cross-sectionally: not just 'is this ratio low' in isolation, but 'is it lower than most of the rest of the stocks scanned in this same batch' - because what counts as 'low' differs enormously between a $2 microcap and a blue chip.

Two guides on Fibonacci trading (Carolyn Boroden's book, and Tickmill's introduction) are in the Resource Library above - here's the honest short version. The idea: after a stock moves from a swing low to a swing high (or vice versa), traders watch specific retracement levels - 23.6%, 38.2%, 50%, 61.8%, 78.6% - as places price is somewhat more likely to pause or reverse on its way back. These numbers come from the Fibonacci sequence's ratios, but the 50% level (the most commonly respected one) has no Fibonacci origin at all - it's just the halfway point, and traders watch it because other traders watch it. Both source documents are refreshingly honest about this: Tickmill's own guide says directly that Fibonacci analysis 'should not be deemed as the magic wand' and works best combined with an actual confirmation signal (a candlestick pattern, a trendline, a moving average) - never traded blind. Fibonacci isn't currently a built feature in Signal H3; see the Dashboard for what it would take to add it (swing-point detection plus simple ratio arithmetic - genuinely buildable with existing data, unlike most of the institutional material referenced elsewhere in this app).

The very first version of Upcoming Top 50's jump-history check came from studying one real example: CGD.V's move from a quiet base into a 793% run. Ordinary volatility measures treat every day's price move the same way. Jump-diffusion math separates a stock's ordinary day-to-day wiggle from genuine SUDDEN jumps, using a technique called bipower variation - the mean product of adjacent absolute returns, which a single huge one-day move doesn't inflate the way it inflates a plain sum of squared returns. Whatever's left over after removing the ordinary wiggle IS the jump contribution. The practical payoff: this app can tell you whether a ticker has a genuine HISTORY of explosive jumps (how many, how big, how often per year) - not just whether today happens to be a big volume day.

The Options Chain tab computes real Greeks (Delta, Gamma, Theta, Vega) locally using the actual Black-Scholes formula - here's what those numbers mean without the math notation. Delta: how much the option's price moves for a $1 move in the stock. A Delta of 0.65 means the option gains roughly 65 cents for every $1 the stock gains. Gamma: how much Delta itself changes as the stock moves - high Gamma means Delta can shift quickly. Theta: how much value the option loses per day just from time passing, all else equal. Vega: how much the option's price changes if the market's expectation of future volatility (implied volatility) shifts by one percentage point. These Greeks are computed from the contract's OWN quoted implied volatility - this app isn't guessing at volatility, it's using whatever the real options market already quotes for that contract.

Two of the simplest, most repeated ideas in the institutional research shared this session: On-Balance Volume (OBV) divergence, and the Volume Z-Score. OBV divergence just means: price is going nowhere, but the volume flowing in is steadily piling up in the buyers' favor. Picture it like this: Price: ----________---- (flat, sideways, boring) OBV: _______///////// (steadily climbing anyway) That gap between a flat price and a climbing OBV line IS the divergence - someone is accumulating shares without pushing the price up (yet). Signal H3's own Accumulation Bias check in Upcoming Top 50 is exactly this idea, just under a plainer name. Volume Z-Score answers a narrower, sharper question: how many standard deviations above NORMAL is today's volume for this specific stock? A Z-score of 4 means today's volume is a genuinely rare event for that ticker, not just 'busier than usual.' Signal H3's Prior Volume Anomaly check on the same tab computes exactly this.

VCP shows up constantly across the trading literature shared this session, under different names - and it's already built into Signal H3's own Volatility Contraction and Range Compression checks. The idea: as a stock's pullbacks get SMALLER and TIGHTER each time, it's a sign sellers are running out of ammunition. Picture three pullbacks getting progressively shallower: Pullback 1: \________/ Pullback 2: \____/ Pullback 3: \__/ <- tightening - sellers exhausted Each pullback is a smaller percentage than the last. That narrowing range is the 'coiled spring' - less and less energy is needed to push the price out of it, in either direction. It's a setup pattern, not a signal by itself - it says 'something's about to happen,' not which way.

Several documents shared this session describe combining 2-3 separate quantitative signals into one weighted 'Accumulation Score,' rather than trusting any single indicator alone - the exact same philosophy behind every scoring engine already in Signal H3 (Pattern_Score, Upcoming Top 50's 10-criteria score, and so on). Relative Strength [||||||||||||||||] 40% weight Accumulation (ADF) [||||||||||||] 30% weight Vol. Contraction [||||||||||||] 30% weight = Final Score Why bother blending instead of just picking the 'best' one indicator? Because every single indicator has false positives - a stock can look tight (VCP) for boring reasons, or show high relative strength just because its whole sector is hot. Requiring MULTIPLE independent signals to agree at once cuts down on being fooled by any one of them alone. This is why Signal H3's own scores are built the same way - a sum of weighted, independently-passing checks, never a single number.

When a fund needs to buy (or sell) millions of shares, dumping the whole order at once would spike the price against them before they're even done buying. Almgren-Chriss is the classic math (a 'calculus of variations' framework) for finding the least-bad way to spread that order out over time. Trade FAST -> big impact now, but done quickly (less risk price drifts away) Trade SLOW -> tiny impact each day, but price might drift away while you wait Almgren-Chriss finds the mathematical sweet spot between these two costs, given how much risk the fund is willing to tolerate. This is genuinely NOT something Signal H3 builds - it solves a problem this app doesn't have, since it never places an order large enough to move a real market; it's paper-trading only. Worth understanding anyway, because it explains WHY a large real buyer's volume often shows up gradually over days, not all at once - which is part of what a 'stealth accumulation' check like Signal H3's is actually trying to detect from the outside.

Ordinary statistics often assume events happen independently of each other. Hawkes processes exist because real order flow doesn't work that way - one big trade tends to trigger a chain reaction of more trades. Big institutional block trade | v triggers -> other algorithms react to the price move | v triggers -> more algorithms follow | v triggers -> retail traders notice and pile in (FOMO) Each event makes the NEXT event more likely for a while, then that extra likelihood decays back down over time - like ripples that fade out. This requires classified buy/sell tick-by-tick trade data to actually fit, which this app doesn't have access to (only daily OHLCV bars) - genuinely not buildable here without a paid tick-data subscription.

VPIN (Volume-Synchronized Probability of Informed Trading) is how market makers try to sense when someone trading against them knows something they don't - before it shows up as a price move. One volume 'bucket' of trades: [ BUY BUY BUY BUY | sell ] 75% buy-side 25% sell-side <- lopsided = suspicious The idea: split trading volume into equal-sized buckets (not equal TIME - equal VOLUME), and measure how lopsided each bucket is between aggressive buying and aggressive selling. A consistently lopsided imbalance is read as a sign that 'informed' (possibly better-informed) traders are pushing one direction - and market makers respond by widening their spreads or pulling their quotes entirely, which is part of why thin micro-caps can suddenly gap with no warning. Needs classified buyer-initiated/seller-initiated trade data this app doesn't have - not buildable here without a paid tick-data feed.

This one's less a technique and more a reality check on WHY the previous few articles (Hawkes, VPIN) can't be built into Signal H3. Retail data (what this app has): One row per day: Date, Open, High, Low, Close, Volume Institutional data (Level 3 / ITCH-OUCH feeds): Every single quote, cancellation, and trade, timestamped to the NANOSECOND, all day long That's a difference of maybe 5-10 numbers a day versus millions of events a day, for the same one stock. On top of the data itself, the fastest institutional desks also place their own servers physically inside the exchange's own data center (co-location) and write trading logic directly onto custom chips (FPGAs) to shave their reaction time down to a few hundred nanoseconds. None of this is something an individual can realistically access - the data alone costs real money from a vendor like Databento or Polygon.io, and true co-location contracts run into the hundreds of thousands of dollars a year. Genuinely out of scope for this app, and honestly for almost anyone outside a funded trading firm.

Richard Wyckoff's accumulation schematic, from the early 1900s, describes a five-phase story for how a stock quietly changes hands from weak holders to strong ones before a real breakout: Phase A: the prior downtrend stops Phase B: a quiet, sideways range builds ('the cause') Phase C: the 'Spring' - a sharp shakeout BELOW support, scares out weak holders Phase D: 'Sign of Strength' - a strong push higher, on clearly rising volume Phase E: Markup - the actual trend move plays out The 'Spring' in Phase C is the part that trips people up - it LOOKS like a breakdown, which is exactly the point: it's designed to shake out anyone without conviction right before the real move. This isn't a single number Signal H3 computes - it's closer to a narrative pattern across several of the app's existing checks at once (volatility contraction building 'the cause,' a volume anomaly for the shakeout, then the dry-up-then-spike check for the actual Sign of Strength).

When you compute how hundreds of stocks move together, most of what you get back is just statistical noise - not real economic relationships. Random Matrix Theory (specifically the Marchenko-Pastur law) gives you a mathematical boundary for exactly how much correlation you'd expect from pure noise alone, so you can throw that part away. Correlation matrix eigenvalues, sorted: [noise][noise][noise][noise][noise] [ REAL SIGNAL ] <----------- discard these -------> <-- keep this --> Anything below the noise boundary is indistinguishable from randomness and should be ignored when building a portfolio; anything above it is a genuine structural relationship worth paying attention to. This is real, honestly buildable math using only the daily price data this app already has - it just doesn't have anywhere to go yet, since the Portfolio tab tracks P&L but doesn't do any correlation-based analysis. The most realistic 'not yet built' item from everything covered this session - flagged here rather than just left out.

A plain random walk has no memory - which direction a stock moved yesterday tells you nothing about which direction it moves today. Real markets aren't quite like that. The Hurst exponent (H) is a single number that measures exactly how much 'memory' a price series actually has. H > 0.5: persistent / trend-following ___/‾‾‾/‾‾‾/___ H = 0.5: pure random walk (no memory) /\/\/\/\/\/\/\ H < 0.5: anti-persistent / mean-reverting \/\/\/\/\/\/ 'Rough volatility' research found something surprising: a stock's own VOLATILITY (not its price) tends to have a Hurst exponent way BELOW 0.5 - meaning volatility itself is unusually jagged and reversal-prone, more than standard models like Black-Scholes assume. This directly explains why options pricing in this app (Options Chain) uses plain Black-Scholes rather than one of these rough-volatility models: Black-Scholes is the right-sized, well-established tool for what this app actually needs, while a full rough-volatility model would be genuine overkill with no exotic-derivatives feature here to justify it.