Documentation

sqzme puts price, volatility, and a few tantalizing traces of market structure next to what happened afterward. The useful question is simple: when the market looked like this before, what followed? The answer is a distribution of historical outcomes. The interpretation is up to the user.

Start with a familiar ticker on the Chart. Look at P, V, G, and D, then try R, I, or F; restrict the dates; then click Nearest. Look at how many past observations remain and how widely their outcomes differ. Change the dates or an axis and see whether the apparent relationship survives. If it evaporates, that is useful information too.

This page explains the measurements, the tools that use them, the fields you can download, and the limits we think you should keep in view. The About page is a shorter statement of temperament. The dated arguments, mistakes, and revisions live in Weekly Notes; subscribers can search the full text there.

Contents

1. Two kinds of chart
2. What goes into a daily row
3. The measurements
4. What the historical forecasts say
5. Using the site
6. Downloads and API
7. What can go wrong
8. Reading and sources

Two kinds of chart

The public DIX and GEX chart is about an S&P 500 aggregate. DIX is an aggregate dark-pool index; GEX is an aggregate estimate of option gamma exposure. The Chart for an individual ticker uses P, V, G, and D. In particular, its G is a ratio of estimated call and put convexity, and its D is that ticker's smoothed dark ratio. Neither is the public chart's dollar GEX or index-level DIX. Similar letters; different measurements.

What goes into a daily row

A ticker's daily file brings together price and volume, reported off-exchange short-sale volume, listed-option information, and implied volatility. Those inputs have different sources and arrival times. DATE names the market session, not a guarantee that every field was available at that session's close. Check the newest published date before treating a row as current.

Coverage differs by ticker and field. A ticker may have plenty of price history and little usable option history. Missing inputs can leave a raw value, a normalized value, or a forecast blank. There is one important exception in the present calculation of D: a missing daily short-volume ratio is filled with a neutral 0.5 before the five-session average and one-session lag are applied. Thus a displayed D need not mean all of its underlying daily observations were present. We would rather tell you that than let a smooth line imply otherwise.

Upstream corrections can revise historical files. If you are testing a strategy, keep the file version or retrieval date you used. A completed historical file is not automatically a record of what was knowable at every earlier decision time. Its forward-outcome fields deserve particular care.

The measurements

P: price trend

Over the most recent 21 trading sessions, take the mean close-to-close percentage return and divide it by the mean absolute close-to-close percentage return over those same sessions:

P = mean(daily return, 21) / mean(abs(daily return), 21)

P asks how much of the movement has pointed in one direction. Positive means the signed moves have leaned upward; negative means downward. Around zero can mean offsetting moves or no durable drift. P is dimensionless and normally between −1 and +1. It is not a prediction of the next month's percentage return. ADM21, the denominator, tells you how large the typical daily move was in percentage points. If that denominator cannot be estimated, P is absent.

V: volatility trend

ADM21 is the mean absolute daily percentage move over 21 sessions. V is ADM21 minus its own 21-session moving average:

V = ADM21 − mean(ADM21, 21)

Positive V says recent realized daily movement is above its recent baseline; negative V says below. V is in percentage points of average daily movement. It is neither annualized volatility nor a forecast, and raw V is not bounded between −1 and +1. V_NORM uses a different scale.

G: gamma ratio

G estimates the call-side share of listed-option convexity for one ticker. For each listed contract, the calculation models the change in delta after a small move in the underlying under a common volatility assumption, then weights that change by open interest. It compares the call-side magnitude with the combined call and put magnitude:

G = modeled call convexity / (modeled call convexity + modeled put convexity)

With usable inputs, G is between 0 and 1. Around 0.5 is roughly balanced; nearer 1 is more call-heavy. It describes an option-position proxy. Open interest counts outstanding contracts, but does not reveal who owns every side or exactly how a dealer is hedged. Our Gamma Exposure paper gives the dealer-hedging hypothesis behind the broader work. The ticker ratio is not the public S&P 500 dollar-GEX series.

D: dark ratio

The raw daily ratio is reported off-exchange short-sale volume divided by its corresponding reported total volume. The current ticker calculation fills a missing ratio with 0.5, averages the resulting series over five sessions, and lags that average by one session before it appears beside the other measurements. The lag keeps the same-day short-volume print out of a state that is meant to be evaluated before that print is available. The neutral fill can mask a missing source day; check coverage if D matters to your analysis.

Some off-exchange short sales can be the other side of a customer's purchase when a liquidity provider supplies shares it does not yet own. Short is Long develops that interpretation. D is still a proxy, not a count of customer buy orders. Daily short-sale volume also is not the same thing as outstanding short interest; FINRA explains the distinction and documents the daily feed.

R: realized vol

R converts the trailing 21-session mean absolute daily percentage move (ADM21) to an estimated annualized standard-deviation scale: R = ADM21 / sqrt(2/π) × sqrt(252). It is displayed in annualized percentage points, like I. The conversion assumes approximately zero-centered, normal daily returns; R is therefore an ADM-based estimate, not the exact sample standard deviation of returns. Missing ADM21 leaves R missing.

I: implied vol and normalized columns

I is the Chart name for the ticker CSV's IV field. It is an option-implied volatility estimate expressed in annualized percentage points. It reflects option pricing, not movement that has already occurred. The calculation can draw on different option-data paths, so treat it as an estimate rather than an exact quote from one contract.

F: fudginess

F is the Chart name for gross customer option delta dollars divided by same-day stock dollar volume. Opposing option trades do not cancel in the gross numerator. It is a modeled Options Flow ratio, not an annualized volatility measure. Missing flow observations remain gaps; the underlying feed and API keep their existing field names.

The ticker CSV's *_NORM columns put P, V, G, D, and IV on a bounded scale. Moon compares each value with up to 252 sessions, uses its median and median absolute deviation, requires at least 126 finite observations, and compresses the score with tanh; IV is transformed in log space first. The Chart derives similarly bounded normalized R and F values from earlier sessions for its interactive Nearest selection. These are not probabilities, and equal numeric values need not have equal economic meaning. A missing value or insufficient history leaves the corresponding normalized reading blank.

What the historical forecasts say

The file's P_NN is an estimate of P 21 trading sessions later. For each eligible historical date, the calculation finds neighbors in the five original normalized measurements (P, V, G, D, IV) and averages the matured future-P values of the nearest one-sixth of eligible rows, about 16.7%. R and F do not enter that published forecast. A recent historical row without its full forward outcome cannot contribute that outcome. P_NN is not a projected percentage price return.

The Chart's interactive Nearest selection is a different operation. It compares the current state with eligible past rows across whichever normalized axes are active on the Chart. One click keeps roughly half the pool; further clicks narrow it. Watch the count as you go. Similarity in these coordinates does not make two episodes identical.

The separate Research view describes a conditional five-day price-return distribution. It reports its analogue count, mean and median, and a 5th-to-95th-percentile interval. Five-day return, 21-session future P, and the Chart's exploratory slices answer different questions. None is an instruction to trade.

Using the site

Chart

The Chart puts a ticker's price and measurements beside historical forward outcomes. It starts with P, V, G, and D. Its seven selectable ticker indicators are P, V, G, D, R, I, and F; R and F appear where their source and normalization history are sufficient. You can change the visible measurements, add peer-ticker P panels, restrict the date range with the context brush, and select points in the indicator scatterplots. The outcome view offers five- and 21-session horizons and different return scales when the needed data exist.

Nearest narrows the candidate history by similarity. Best and Worst search the upper or lower half of a displayed measurement for the strongest or weakest median forward outcome within the current pool. That is a way to find a question, not an out-of-sample performance result. The more slices you try, the easier it is to find an obliging one by chance. Reset clears exploratory restrictions.

Ideas

Ideas scans tickers by the 21-session forecast of P. It groups positive-or-flat, negative, and unavailable forecasts. Search by ticker or company; choose stocks, ETFs, or all tickers; set the usable-history threshold; and sort by forecast strength, direction, history, or ticker. The default 252-session threshold screens for length, not forecast quality. Open a ticker's Chart before you decide what a ranked row means.

Research (deprecated)

The separate Research distribution is specific to a ticker; here is SPY as an example. It shows the current state against historical analogues, a conditional five-day return range, and a scenario view; its model record can be downloaded as JSON. An elegant curve drawn through a thin or old sample is still a thin or old sample. Inspect the analogue count and dates. This endpoint is no longer fully supported, but some of you people still love it.

Options Flow and P-scope

Options Flow is a separate daily feed of estimated option-trading activity. The Chart offers its gross-delta-to-stock-volume ratio as F. Other flow measurements remain in the API file. Trade classification and greeks are modeled, not direct observations of customer positions. Missing flow dates remain gaps; a missing stock-volume denominator makes F unavailable. Check the feed date and classified-volume fraction before leaning on a spike. The API Explorer exposes the flow file and status.

P-scope explores lead-and-lag relationships among tickers. As with Chart selections, keep the dates and the path you took to a pleasing pattern in view.

The P-scope has been rolled into the Chart page. Use the (+) button at the bottom of the page to add a peer ticker to the panel or to use suggested peers.

Weekly Notes

The Yacht Club keeps the dated Weekly Notes. Its archive cards sit beneath the timeline, and search looks through the full note text rather than only card excerpts. The notes are where we argued for these measurements, changed our minds, and sometimes found that an edge had gotten away from us. They are part of the record; this page is the current reference for what the product actually calculates.

Downloads and API

The Chart exports a ticker's daily CSV. The API Explorer builds read-only GET requests for a ticker (/symbol/{ticker}), the latest cross-section (/latest), Options Flow by ticker and its status, and available Yacht Club files. Symbol and latest data can be CSV or JSON; a symbol request can join Options Flow columns by date with the prefix OPTIONS_FLOW_.

The base path is https://squeezemetrics.com/monitor/api. All requests use an API key in the key query parameter. The Explorer supplies the exact request shape and available formats.

Create or rotate an API key on Profile. The Explorer places the key in the generated URL. Treat that URL as a credential: keep it out of screenshots, shared notebooks, repository commits, and links that may expose their address. Copy a current request from the Explorer rather than a sample key from documentation. A missing or invalid key, or an account without access, will not return subscriber data. Our subscribers are well-behaved and reasonably clever, so API keys will be rotated aggressively in the event of apparent abuse.

R_21F and P_21F are outcome labels. They are blank at the recent edge because the future has not happened yet. For a backtest, keep them out of the inputs at a historical decision date, wait for the full outcome period to mature, and account for later source corrections. The Explorer remains the exact guide to live endpoints and request shapes.

What can go wrong

• A proxy is taken literally. D is not customer intent counted directly. G is not a view into each dealer's book. Options Flow is classified and modeled.
• A neat line hides missing data. D's neutral fill is one example. A missing normalized value or forecast is another kind of warning. Check the dates and coverage underneath the picture.
• The sample is mistaken for independent evidence. Neighboring 21-session outcomes overlap. One event can supply several nearby rows. A longer history may include an unlike market or an unlike version of the ticker.
• The best-looking slice was selected after looking. Chart's Best and Worst, several Nearest clicks, or many manual brushes can generate a persuasive in-sample picture. Try a different window, inspect the whole distribution, and leave a period untouched for a real test.
• An old relationship is treated as a law. Market participants change their behavior. What worked as a signal can decay as it is noticed, traded, or simply overtaken by a different market structure.

The practical habit is to look at dates, counts, raw measurements, and the whole range of outcomes before narrating the median. If a relationship survives those checks, you have a better question, not a guarantee.

Reading and sources

The three foundational papers are Short is Long, Gamma Exposure, and The Implied Order Book. They show where several ideas came from; they are dated research rather than a warranty for today's measurements. For source definitions, see FINRA's short-volume guide, its short-volume/short-interest distinction, and OCC's open-interest reports.

For ideas as they changed, see the dated notes on early GEX assumptions, whether DIX had lost predictive power, and signals in an adaptive market. Follow those into the searchable Weekly Notes archive rather than treating any one note as the final word.

For the current offer, see Plans; for your own account, see Billing. If a definition here does not reconcile with a downloaded row, write to support@sqzme.co with the ticker, date, field, and what you expected.