The Game Theory Panel, in plain English

What every number means, what it is actually telling you, and what to do about it. No maths required — the maths is in the code if you ever want it.

START HEREThe one idea behind all of it

A normal indicator looks at price and asks "what shape is this making?" Every tool on this panel asks a different question: who is on the other side of this trade, and what are they going to do next?

That sounds soft, but it isn't. There is fifty years of published research on exactly this, and it gives you numbers. When you buy, someone sold to you. That person is either a market maker who wants to be flat by tonight, another punter guessing like you, or someone who knows something you don't. Those three counterparties behave completely differently, and you can tell them apart from price and volume alone.

That's what the panel does. Every box on it is answering one part of the same question.

If you read nothing else: the score at the top tells you direction and conviction. The four tiles below tell you whether the market will actually let you collect it. A great-looking signal in a toxic, thin, crowded tape is not a great trade.

SECTION 1Game Theory Score — the headline number

What it is

One number from −100 to +100 that blends everything else on the page. Positive means the weight of evidence is bullish, negative bearish. The size matters as much as the sign.

How to read it

Reading	Means	What you do
+35 and above	Long signal	Take the trade. Size it by how far above 35 it is.
−35 to +35	No edge	Do nothing. This is most of the time, and that's correct.
−35 and below	Short signal	Take the short, or if you're long-only, get out.

Think of it as a dial, not a switch. A score of +38 and a score of +85 are both "long", but the first is a toe in the water and the second is a real position. The panel shows this as "conviction" — that's just how big the number is.

Worked example — the panel's "trending tape" demo

The three direction signals came in at +66.2 (order flow), +92.2 (overhang) and +86.1 (trend × regime). Weighted together that's a raw direction of +81.0 — about as bullish as this thing gets.

Then reality gets applied. Flow was toxic (VPIN in the 81st percentile), so the score was cut by 19% → +65.6. Then the crowding inversion flag fired and knocked off a further 25 points → +40.8.

So: still a long, but roughly half the position the price action alone would have argued for. That gap between "+81 on the chart" and "+40.8 after accounting for who's on the other side" is the entire point of the tool.

SECTION 2What is driving it — the three bars

This box breaks the headline score into the three things pushing it around, so you can see why it says what it says. Never trade the composite without glancing at this.

Order-flow imbalance

Plain version: over the recent window, was more volume hitting the offer (buying) or the bid (selling)?

You can't see this directly on a chart — the exchange doesn't tell you who initiated each trade. But there's a well-established trick: a bar that moved up a lot relative to its normal volatility was almost certainly buyer-driven. Do that for every bar, add it up, and you get net demand. Positive = buyers leaning on it.

Capital gains overhang

Plain version: is the average person holding this thing sitting on a profit or a loss?

This one is genuinely underrated and I'd pay attention to it. More in Section 6.

Trend × regime

Plain version: how stretched is price, multiplied by whether stretch is currently a reason to follow or a reason to fade.

This one confuses people — read this bit

This bar can look "wrong" and be right. In the mean-reverting demo tape, price sat below its average, and the regime test said "fade extremes". Below-average price × fade-the-move = buy. So the bar showed +88, bullish, even though price was falling.

That's not a bug. In a mean-reverting market, weakness is the buy signal. The indicator has already done that flip for you — which is exactly why it isn't just a moving average.

Risk multiplier (the purple note underneath)

A number between 0.40 and 1.00 that the whole score gets multiplied by. 1.00 = no penalty. 0.70 = the score was cut by 30% because conditions are bad.

Crucially, it can never flip the sign — only shrink toward zero. If the market is dangerous, the right answer is to trade smaller, not to trade the opposite way. That's a deliberate design rule, not an accident.

SECTION 3VPIN — is the flow toxic?

The plain version

Imagine you're the guy making a market — quoting a price to buy and sell all day, trying to earn the spread. Your nightmare is trading against someone who knows something. You can't tell who's who, so you watch for a tell: is the flow coming at me lopsided?

Normal, healthy flow is two-way — some buying, some selling, roughly balanced. When it goes one-way and stays one-way, someone is working a big order and they probably know why. So you widen your prices, or you stop quoting entirely.

That withdrawal is what causes air pockets. Not the selling itself — the disappearance of the people who normally catch it.

How to read it

Reading	Means	What you do
90+	Toxic	Don't fade the move. Don't sell options into it. Widen stops or stand aside — gap risk is elevated.
20–90	Normal	Business as usual.
Under 20	Benign	Noise traders dominate. Best conditions for mean-reversion and for being the one providing liquidity.

The one-line takeaway: high VPIN doesn't tell you which way it's going. It tells you the move that's coming will be disorderly.

SECTION 4Kyle's λ — how thin is the ice?

The plain version

How much does the price move when someone leans on it with a given amount of money? That's it. That's lambda.

Deep, liquid market: you can push a lot of size through and barely move it. Thin market: the same order moves it 2%. The number here is a percentile — where today's price impact sits against this instrument's own history.

How to read it

Reading	Means	What you do
80+	Thin book	Cut your position size. Your own entry and exit will cost you. Expect gaps and slippage.
20–80	Normal	Trade your usual size.
Under 20	Deep book	You can size up. Good conditions for larger positions.

The clever bit: "stealth accumulation"

The Pine version flags something specific — lambda rising while price goes nowhere. That combination means liquidity is quietly being pulled while the chart looks boring. Someone is accumulating, and the market makers have noticed before the chart has. It's one of the few genuinely forward-looking signals here.

Also check: the R² number

In the detail table you'll see Kyle regression R². This is the honesty check. It asks: is order flow actually what's moving this price right now?

Above 0.10 — yes, lambda is meaningful.
Below 0.10 — no. Price is being moved by something else (news gap, index rebalance, an earnings print). Ignore lambda today.

Most indicators would hide this. This one tells you when it doesn't know.

SECTION 5Regime — which game is being played?

The plain version

Some markets reward you for joining the move. Others punish you for it and pay you for fading. Run a trend strategy in the wrong one and you will lose money consistently while doing everything else right.

This box tells you which one you're in — and, importantly, whether it's sure.

How to read it

Reading	Means	What you do
VR above 1, z above +1.96	Trending. Moves keep going.	Follow the trend. Breakouts work. Buy strength.
VR below 1, z below −1.96	Mean-reverting. Moves get given back.	Fade extremes. Sell rips, buy dips. Breakouts fail.
z between −1.96 and +1.96	Random walk	Stand down. There's no structure to exploit.

Why the z-score matters more than the VR

Almost every "regime indicator" you'll find online gives you a number and lets you believe it. This one gives you a statistical significance test alongside it.

A VR of 1.15 sounds trending. But if the z-score is 0.8, that reading is indistinguishable from random noise — you'd get it by chance on coin flips. The shaded band on the chart is exactly that: inside the band, there is no regime, whatever the line looks like.

Most of the money lost by systematic traders is lost trading a regime that wasn't statistically there. This is the single most useful box on the panel and it's the one people will skip.

SECTION 6Cost basis & overhang — where is everyone's break-even?

The plain version

The purple dashed line is roughly what the average holder paid. Not a moving average — a volume-weighted estimate of where shares actually changed hands.

Why that matters comes from behavioural finance, and it's one of the most replicated findings in the field: people sell winners too early and hold losers too long. They hate crystallising a loss, and they love booking a small gain.

So the gap between price and that line tells you where the selling pressure lives:

Situation	What's happening	What you do
Price well above the line	Most holders are in profit and will take some off into strength. So good news gets underreacted to at first — then the price drifts up as that supply clears.	This is a momentum setup. Expect a drift, not a spike.
Price well below the line	Most holders are underwater and waiting to "get out at break-even". That line becomes hard overhead resistance.	Expect rallies to stall at the purple line. Don't buy breakouts into it.
Price sitting on the line	The average holder is flat. No behavioural pressure either way.	Genuine battleground. Whoever wins this level usually gets a real move.

Worked example — the "blow-off" demo tape

Price 566, cost basis line 367. Overhang +35%. The average holder is up 35%.

Read that two ways at once, because both are true: there's a big cushion of profitable holders who'll support dips — and a big pile of latent supply that will hit the market the moment sentiment cracks.

Overhang half-life

Roughly how many bars until the current crowd of holders has substantially turned over and been replaced. Useful for one thing: sizing your holding period. If the half-life is 80 days, a drift-based trade off this signal is a multi-month idea, not a two-week one.

SECTION 7Crowding — how full is the bar?

The plain version

There's a famous problem in economics called the El Farol Bar. A bar is fun if fewer than 60% of people show up, and miserable if it's packed. Everyone decides independently whether to go. The punchline: if everyone uses the same good reasoning, they all show up on the same night and they all have a bad time.

Markets are the same. A trade doesn't stop working because the thesis was wrong. It stops working because everybody found it. The edge gets destroyed by being discovered.

This box scores 0–100 for how full the trade looks.

How to read it

Reading	Means	What you do
Under 40	Room to run	Fine to build a position and add to it.
60–80	Crowded	Stop adding. Stop pyramiding. Tighten your stop. Do not short it.
80+	Extreme	Take some off. You're being paid to be patient now, not aggressive.
"Inversion" flag	Crowded and it's stopped working	This is the actual exit signal. Reduce.

The mistake nearly everyone makes here

Crowded is not a sell signal. Crowded trades stay crowded and keep paying, sometimes for a very long time. Shorting something purely because it looks extended is how people get run over.

What you're waiting for is the inversion — crowding still high, but the payoff to being in the crowd has started decaying. Crowd's all in; trade's stopped working. That's the signal.

Worked example — two demo tapes, opposite conclusions

Blow-off tape: crowding 90 — extreme. But the inversion flag is off, because it's still going up. Score stays +69.5. Verdict: stay long, stop adding, trail your stop.

Trending tape: crowding only 74 — less extreme. But inversion is ON: it's crowded and momentum is rolling over. So it takes a 25-point penalty and the score drops to +40.8. Verdict: still long, but this one is losing its grip.

The less-extended one is the more dangerous one. That's the whole insight.

SECTION 8Execution planner — don't be your own worst enemy

The plain version

Here's the thing people forget: your order is order flow too. The market maker prices your buying exactly the same way he'd price an insider's. He doesn't know you're not one.

So if you buy a big slug in one go, you pay for the privilege. The cost follows a well-established rule: impact scales with the square root of your size relative to daily volume. Four times the size costs twice the impact.

The flip side is the useful part — spreading an order over more days cuts the total cost by the square root of the days. Patience is worth real money, and this box tells you how much.

How to use it

Type in four things: how many shares, the average daily volume, the price, and the daily volatility as a percentage. It tells you what it'll cost done in one go, and what to do instead.

Worked example — the default numbers

250,000 shares · 1.5m average daily volume · $48.20 · 2.5% daily vol

That's 16.7% of a day's volume. Done in one day it costs about 102 bps — roughly $123,000 in market impact.

Spread over 16.7 days at 15,000 shares a day, the same order costs about 25 bps — roughly $30,100.

Same trade. $92,900 difference. That's just pacing. On a position that size, execution is a bigger lever than most people's entry signal.

Rules of thumb: stay under ~10% of daily volume per day. Above 50% of daily volume, you are the informed trader in the model and the market will price you as one — at that size, rethink the position, not the execution.

SECTION 9Putting it together — how I'd actually use it

The signals aren't a checklist to average. They're a sequence, and they answer different questions in a specific order.

Is there a game to play at all? Check the regime z-score first. If it's inside ±1.96, stop. There's no structure to trade. Go and do something else — this is the discipline most people skip and it's where most of the losses are.
Which direction? Now look at the score and the three bars behind it. Is it order flow driving it, the overhang, or the trend? If one bar is carrying the whole thing on its own, be more sceptical than if all three agree.
Will the market let me collect? Check VPIN and Kyle's λ. Toxic flow or a thin book doesn't mean don't trade — it means trade smaller. The risk multiplier has already done some of this for you; these tiles tell you why.
Am I early or late? Check crowding. Under 40 you can build and add. Over 60, take what's there and stop adding. If the inversion flag is on, you're late — reduce.
Where's the line in the sand? Look at the cost basis chart. That purple line is where behaviour changes. It's a far better stop location than a round number or a moving average, because real people have real money at that price.
How do I get in without paying for it? Run the execution planner. If the position is meaningful relative to daily volume, work it over days. This is free money and almost nobody bothers.

The trade you actually want is: significant regime, score above 35, VPIN under 80, crowding under 60, no inversion flag, and price near the cost-basis line. That combination isn't common — which is the point. Most days the honest answer is "no trade", and a tool that says so is worth more than one that always has an opinion.

SECTION 10The TradingView scripts

Seven scripts, completely standalone from the website. Each is the deep version of one panel section, plotted on your own charts with alerts.

Script	Use it when
`01` VPIN toxicity	You want to know if a move is orderly or about to gap.
`02` Kyle's λ	Position sizing, and spotting quiet accumulation.
`03` Regime	Deciding whether to run trend or mean-reversion this month. Start here.
`04` Crowding	Deciding whether to add to a winner or take some off.
`05` Cost basis	Finding real support and resistance. Great as a standalone overlay.
`06` Game Theory Score	The composite, on a chart, with alerts.
`07` Strategy	Backtesting the composite.

All of them detect whether you're on daily equity bars or intraday crypto/futures bars and retune themselves. You can force it in the Mode setting.

If you backtest script 07, please read this

It has a lot of settings, which makes it easy to fool yourself. Leave the commission and slippage on — a frictionless backtest is a fantasy. Want at least ~50 trades before believing anything. Treat a profit factor above 2.5 as a red flag, not a triumph. And change one setting at a time — if small tweaks swing the equity curve wildly, the edge isn't real.

SECTION 11What this will not tell you

Worth being straight about the limits, because a tool you trust blindly is worse than no tool.

It doesn't know who's actually buying. Price and volume data never says who initiated a trade. The buy/sell split is a well-established inference, not an observation. With real tick data you'd do better.
It needs history. Under about 250 bars the percentile readings are guesswork. Don't run it on a recent IPO and take the numbers seriously.
VPIN here is the simplified version. The proper one runs on a volume clock rather than a time clock. On liquid instruments they track each other closely; on daily bars, read it as slow distribution pressure rather than an HFT signal.
It describes the present, it doesn't forecast. "Toxic flow" means conditions are dangerous now — not that a crash is coming.
It can't see the news. An earnings miss or a takeover bid overrides everything on this page instantly.
None of this is investment advice, and you already knew that.

Every model here comes from published, peer-reviewed research — Kyle (1985) on price impact, Glosten-Milgrom (1985) on adverse selection, Lo-MacKinlay (1988) on regime testing, Grinblatt-Han (2005) on the overhang effect, Easley-López de Prado-O'Hara (2012) on flow toxicity, Arthur (1994) and Challet-Zhang (1997) on crowding, and Almgren-Chriss (2001) on execution. Full citations are in each script's header and in the README.