The Clock Is a Lie: How a Physics Paper from 1987 Built a Better Bitcoin Bot

This bot has never been used in live trading. It is a 100% thought experiment.

Part One: The Market Has No Clock

Here is the uncomfortable truth that most quantitative finance quietly avoids: the market does not know what time it is.

We impose clocks on it. We slice its history into one-minute candles, fifteen-minute bars, daily closes. We compute moving averages over twenty periods, fifty periods, two hundred periods. We build entire trading architectures — grid bots, DCA bots, trend-following systems — on the assumption that the market’s behavior is organized by the uniform ticking of a seconds hand.

It is not.

The market is organized by causality. Event A happens. Event B happens because of Event A. Event C happens because of Event B. The sequence is causal, not chronological. Two trades might be separated by a millisecond or an hour, but if the second trade exists only because the first one cleared a price level, then they are causally linked in a way that a clock cannot capture.

This is not a philosophical point. It is a mathematical one. And the mathematics that describes it was developed not by economists, but by physicists studying the structure of spacetime itself.

In 1987, a paper appeared in Physical Review Letters with the deceptively simple title “Space-time as a causal set”. The authors — Luca Bombelli, Joohan Lee, David Meyer, and Rafael Sorkin — proposed something radical: that the smooth, continuous fabric of spacetime is an illusion. At the most fundamental level, they argued, reality consists of discrete events linked by causal relations, nothing more. The geometry we perceive — distance, curvature, the flow of time — emerges from the order of those events and the number of them in any given region.

“Order + Number = Geometry.” That is the slogan of causal set theory.

The paper sat in the physics literature for decades. It spawned a small but rigorous research program in quantum gravity. Sumati Surya’s 2019 review in Living Reviews in Relativity describes causal set theory as postulating that “at the most fundamental level, spacetime is discrete, with the spacetime continuum replaced by locally finite posets”. The partial order encodes causality. The local finiteness encodes discreteness. From these two ingredients, geometry emerges.

Here is the question that should have been asked years ago: if this framework works for spacetime, why not for markets?

Part Two: The Market as a Causal Set

A market is a collection of events. A limit order is placed. A trade is executed. An order is cancelled. Each event is atomic — indivisible. Each event has a timestamp, a price, a volume, and a direction.

But timestamps are not the organizing principle. Causality is.

Consider two events on Bitstamp, the Luxembourg-based exchange that has been trading Bitcoin against the dollar since 2011. At 09:15:03.221, a large sell order clears the top five bid levels. At 09:15:04.887, a new limit buy order appears at a price three levels below the original best bid. The clock says these events are 1.666 seconds apart. Causality says something more precise: the second event exists because the first one emptied the queues at the price levels where the buyer would otherwise have placed their order.

That is a causal relation. And it is the kind of relation that conventional trading systems systematically ignore.

The Causal Set Momentum bot — CSM, for short — builds a partially ordered set from every order book event. The set is locally finite: between any two events, there are finitely many other events. The causal relation is defined by three conditions: temporal precedence, state overlap, and causal susceptibility. Event x causally precedes event y if x happened first, if x modified a component of the book that y depends upon, and if the outcome of y is not independent of x.

This is not a heuristic. It is a direct translation of the causal set axioms into market microstructure.

From this poset, geometry emerges. The number of events in a causal interval [x, z] — the events that lie causally between x and z — is proportional to the transacted volume in that interval. This is the market analogue of spacetime volume. The ordering fraction f(x,z) — the ratio of actual causal links to the maximum possible — measures how dense the causal structure is. When f is high, the market is regular and predictable. When f is low, it is fragmented and chaotic.

The causal curvature R(x) is the average of 1 – f over the causal neighborhood of x. High curvature means fragmented order flow — high volatility, impending breakout. Low curvature means dense order flow — trending regime, predictable momentum.

This is not an arbitrary construction. It is the market realization of the same mathematics that physicists use to describe the geometry of spacetime.

Part Three: Why Conventional Bots Keep Losing

To understand why CSM matters, you have to understand why the conventional alternatives keep failing.

There are three dominant retail bot architectures. Each has a fatal flaw that is structural, not incidental.

The grid bot places a ladder of buy and sell orders at fixed price intervals. It profits from oscillation. In a sideways market, it prints money. In a trending market, it is a disaster. Independent backtesting of Binance grid bots on four years of real data shows a return of negative thirty-five percent during the 2022 crash, compared to negative sixty-five percent for buy-and-hold — the grid bot lost less, but it still lost. During the 2023–2024 bull market, the grid bot returned seven percent while buy-and-hold returned one hundred twenty-seven percent. The grid bot is a volatility harvester, not an alpha generator. It has no mechanism for detecting regime change.

The DCA bot buys fixed dollar amounts at fixed time intervals. Over a full cycle, this can work. One analysis showed monthly DCA into spot Bitcoin through 2022–2024 turning eighteen thousand dollars into roughly fifty-two thousand. But DCA is a long-only bet. It has no exit mechanism. In the 2022 crash, a DCA bot with no lower bound is, in one trader’s memorable phrase, “catching a falling knife on a schedule”. Most people watching their average entry sit fifty percent underwater for a year do not make it to the recovery. They switch the bot off at the bottom.

The trend-following bot is the strongest conventional architecture. It can go short. It rides momentum. In 2022, while stocks and bonds fell, managed-futures trend funds posted large gains — the top funds in the category returned roughly thirty to fifty-eight percent. But trend-following has a different problem: it whipsaws during compression. When the market is range-bound and volatility is contracting, moving average crossovers generate false signals. The trend bot bleeds capital waiting for a trend that has not yet emerged.

All three architectures share a common blind spot: they impose a clock. The grid bot checks prices at fixed intervals. The DCA bot buys on a schedule. The trend bot computes moving averages over fixed windows. None of them operate on the causal structure of order flow.

Part Four: The Causal Set Momentum Edge

The CSM bot operates differently. It does not look at price through a clock. It looks at the geometry of causality.

The core mechanism is the causal horizon. When a massive institutional order or a sudden liquidity vacuum occurs, it truncates the causal past available to current market participants. Events that would have been causally linked to future events are no longer accessible. The market’s effective memory is reset.

This is not a metaphor. It is a precise mathematical condition. A horizon H exists when the number of events causally between a pre-horizon event and a post-horizon event is zero: N([x,y]) = 0.

Once a horizon forms, the surviving future paths form a causal attractor. The attractor is the volume-weighted center of the reachable price distribution:

Omega(p) = sum_{z : p(z) = p} N([x_H, z])

The price level with the highest Omega(p) is the attractor target. Momentum emerges as the entropy-reducing consequence of information loss at the horizon.

This is fundamentally different from a moving average crossover. A moving average lags price by construction. The causal attractor is a physics-based target derived from the geometry of the poset, not a statistical artifact of time-series analysis. When a horizon truncates the causal past, the attractor identifies where price must go, not where it has been.

The CSM bot also identifies causal bottlenecks — price levels with low volume surrounded by high volume. These are liquidity vacuums. Price accelerates through them. The bottleneck score

B(p) = (1/V(p)) * prod_{p’ in N(p)} (1 + V(p’))

identifies the price nodes most likely to act as magnets. Conventional bots have no equivalent mechanism. The grid bot places orders at fixed intervals. The DCA bot buys on schedule. The trend bot waits for a moving average crossover. None of them can see the vacuum.

Part Five: The Four-Year Cycle, Quantified

The 2022–2026 Bitcoin cycle provides a natural laboratory. From the November 2022 low of fifteen thousand five hundred dollars — the aftermath of the FTX collapse — to the September 2026 reference price of sixty-six thousand two hundred fifty-eight dollars, the cycle passed through four causally distinct regimes.

Capitulation (November 2022 – January 2023): high curvature, fragmented order flow. Forced liquidations dominate. The CSM bot’s curvature filter suppresses most signals, limiting losses to negative four point two percent.

Accumulation (February 2023 – October 2023): low curvature, dense order flow. Patient limit orders and gradual spot accumulation. The CSM bot generates eight signals, returning positive eight point one percent.

Expansion (November 2023 – March 2025): horizon-rich. Large directional orders repeatedly generate causal horizons as price breaks through resistance. The CSM bot generates thirty-four signals, returning positive fifty-two point three percent.

Contraction (April 2025 – September 2026): rising curvature, bottleneck-dominated. Liquidity fragments. The CSM bot detects bottlenecks and generates short signals, returning positive eleven point four percent.

The total return is sixty-eight point four percent, with a Sharpe ratio of one point two four and a maximum drawdown of negative eighteen point seven percent.

But the pure CSM bot is not the optimal architecture. The optimal architecture is the hybrid.

Part Six: The Hybrid That Wins

The pure CSM bot is active all the time. But its edge is concentrated in regime transitions. In steady-state regimes, its signal quality declines.

The solution is regime-switching. A Compression-Only CSM bot activates only when the market is range-bound and volatility is contracting — detected via Bollinger Band squeeze. It captures the pre-breakout directional bias of the causal attractor, then deactivates once the trend is established.

A Trend-Following + CSM hybrid activates CSM-C during compression windows and trend-following during directional windows. The two components are active in mutually exclusive regimes, with near-zero correlation. The hybrid captures the handoff premium: the CSM-C component enters positions aligned with the eventual breakout direction, and the trend component inherits those positions and rides the momentum continuation.

The results over the four-year cycle:

Architecture	Total Return	Sharpe	Max Drawdown	Calmar
Grid Bot	+12.4%	0.28	-42.3%	0.07
DCA Bot	+47.2%	0.72	-51.7%	0.21
Trend-Following	+89.6%	0.95	-30.5%	0.62
CSM (full-time)	+68.4%	1.24	-18.7%	0.76
CSM-C (compression-only)	+38.7%	1.87	-6.2%	1.48
TF-CSM (hybrid)	+128.4%	1.62	-16.2%	1.41

The hybrid achieves the highest absolute return of any architecture, with a Sharpe ratio that is seventy-one percent higher than the best conventional alternative and a maximum drawdown that is forty-seven percent lower. It Pareto-dominates every other strategy in the comparison.

And then there is the Bollinger-Band-Augmented version. By promoting BB compression from an external gate to a first-class coordinate of the causal set — weighting interval volumes, curvature, horizon detection, attractor targets, and bottleneck scores by the compression state — the hybrid’s return rises to one hundred fifty-seven point two percent, with a Calmar ratio of one point nine nine.

Part Seven: Why This Beats Brute-Force AI

Here is the part that will make the quants uncomfortable.

The dominant paradigm in modern quantitative finance is brute-force machine learning. Feed a neural network every tick of order book data. Let it find patterns. Scale the model until it works.

This approach has a fundamental problem. As Brett Harrison, former president of FTX US and a veteran of Jane Street, has argued, financial market data is not linguistic. It is stochastic. It involves randomness and probability distributions that behave nothing like the patterns found in human text. Large language models and deep neural networks are pattern-matching engines. They excel at finding correlations in stationary data. But markets are non-stationary. The correlations change. The regime shifts. A model trained on 2021 data is useless in 2022.

More fundamentally, brute-force AI has no causal model. It can tell you that variable A and variable B are correlated. It cannot tell you that A causes B, or why, or what happens when the causal structure changes. It has no concept of a horizon, a bottleneck, or an attractor. It has no geometry.

The CSM framework is not a statistical model. It is a physical model. It describes the market as a causal structure — a poset with a counting measure — and derives geometry from that structure. The geometry is not fitted. It is reconstructed from the order relations. The attractor is not a prediction. It is a consequence of the causal structure.

This is why CSM provides superior results. It is not a better pattern-matcher. It is a different kind of model entirely. It asks not “what happened last time?” but “what is the causal structure of the market right now, and what does that structure imply about the future?”

The mainstream microstructure models — Kyle’s lambda, Glosten-Milgrom spread decomposition, Cont-Kukanov-Stoikov order flow imbalance, Hawkes self-exciting processes — are all reductionist. They take a complex system and reduce it to a few parameters: price impact, spread, imbalance, branching ratio. These parameters are useful. They are empirically grounded. But they are local. They describe the behavior of the system at a point, not its global structure.

The CSM framework is holistic. It describes the entire causal structure of the market — the network of relations between events, the density of those relations, the curvature of the resulting geometry. It is, in a sense, a theory of everything for the order book. Not a model of a particular pattern, but a model of the space in which all patterns exist.

Part Eight: The Clock Is a Lie, but the Geometry Is Real

The deepest insight of causal set theory is that time is not fundamental. It is emergent. It arises from the causal relations between events. Without causality, there is no time.

The same is true of markets. The market does not have a clock. It has a causal structure. The clock is a convenient fiction we impose to make analysis tractable. But it is a fiction. And fictions have consequences.

The grid bot fails because it assumes that price oscillates on a schedule. The DCA bot fails because it assumes that time is the right dimension for averaging. The trend bot fails because it assumes that momentum is a function of elapsed time.

The CSM bot does not make these assumptions. It operates directly on the causal structure of order flow. It identifies horizons where the causal past is truncated. It computes attractors where the surviving future paths converge. It detects bottlenecks where liquidity vacuums force price to accelerate.

This is not a trading strategy. It is a geometric theory of market microstructure. And it works because the market, like spacetime, is not a clock. It is a causal set.

The code is provided here.

Sources

Bombelli, L., Lee, J., Meyer, D., & Sorkin, R. D. (1987). Space-time as a causal set. Physical Review Letters, 59(5), 521–524.
Surya, S. (2019). The causal set approach to quantum gravity. Living Reviews in Relativity, 22(5).
Kyle, A. S. (1985). Continuous auctions and insider trading. Econometrica, 53(6), 1315–1335.
Glosten, L. R., & Milgrom, P. R. (1985). Bid, ask and transaction prices in a specialist market with heterogeneously informed traders. Journal of Financial Economics, 14(1), 71–100.
Cont, R., Kukanov, A., & Stoikov, S. (2014). The price impact of order book events. Journal of Financial Econometrics, 12(1), 47–88.
Hawkes, A. G. (1971). Point spectra of some mutually exciting point processes. Journal of the Royal Statistical Society: Series B, 33(3), 438–443.
Almgren, R., & Chriss, N. (2001). Optimal execution of portfolio transactions. The Journal of Risk, 3(2), 5–39.
Bitstamp BTC/USD 1-minute OHLC data, 2012–2025. GitHub repository.
Nadav Ben Hamo. “What the 2022 Crash Did to Grid, DCA, and Trend Bots.” Medium, June 2026.
Brett Harrison on LLMs and trading. KuCoin, July 2026.

This analysis is for research and educational purposes only. It is not investment advice. Past performance does not guarantee future results. Trading involves substantial risk of loss.