Decision Fatigue on the Desk: Why Your Worst Trades Cluster Before the Close

Trading is a decisions-per-hour job

Decision fatigue is the measurable decline in decision quality after a long run of decisions, and trading is its purest expression in professional work. A discretionary trader makes hundreds of discrete judgments in a session - size, entry, exit, hedge, hold, cut - each against a moving price, on incomplete information, with an immediate consequence. Most jobs measure workload in hours. On a desk the measure that predicts when judgment slips is decisions per hour.

The mechanism is specific. Sustained deliberate choice loads the lateral prefrontal cortex, which holds the position in working memory, weighs alternatives and suppresses the impulse to act early. That work runs on glutamate, the brain's primary excitatory neurotransmitter. Every decision fires glutamate and the brain clears it continuously; when decisions arrive faster than clearance keeps pace, glutamate accumulates, neural signalling becomes less efficient, and the output of that circuit degrades. My own research has concentrated on that balance rather than on willpower. The detail is on the Numin science page and in our explainer on glutamate and decision fatigue.

Three properties of desk work stack on that mechanism:

• Density - the decisions arrive in clusters measured in seconds, with gaps too short for clearance to catch up. We call this variable decision density, and trading sits at the top of the distribution.
• Time pressure - a market decision cannot be postponed to a calmer moment. Sleeping on it, the way most professionals ration a depleted decision system, is unavailable while a position is live.
• Immediate consequence - outcomes are marked to market continuously, so each judgment carries an emotional charge that must be suppressed to make the next one cleanly, using the same circuitry as the decision.

Note what is absent: effort, discipline and hours worked. The depletion is physiological, and a depleted judgment feels like a reasonable one from the inside.

The pre-close window is the worst intersection of the day

Put the depletion curve and the trading day on the same axis. In our placebo-controlled trial the placebo arm declined significantly in cognitive efficiency after roughly four hours of sustained load - and for a trader engaged from the pre-market open, four hours is done before lunch.

The final hour is where two curves cross in the worst possible way. Decision capacity is at its daily low - past the four-hour mark, past the post-lunch dip we describe in neuro-metabolic traffic jams, and several hundred decisions deep. Meanwhile demand on it spikes: positions squared or deliberately carried, overnight gap risk sized, end-of-day limits respected, and everything unresolved arriving at once.

"The close does not ask whether your decision system is ready. It arrives on schedule and takes whatever judgment is left."

There is a second reason the window is dangerous. Deferral is the cheapest way to cope with a depleted decision system, and the closing bell removes it. An executive who runs out of judgment at 4pm can push the call to tomorrow; a trader carrying a position into the close cannot. Not deciding is itself a decision with a P&L attached, so depletion cannot express itself as delay. It comes out in the calibration of the decision rather than its timing.

Depletion moves your risk criterion, not just your reaction time

Decision fatigue is usually described as decisions getting slower and less accurate. On a desk that undersells it. The more consequential change is that the criterion moves: what registers as acceptable risk at hour seven is not what registered at hour one, and the shift happens without any conscious revision of the strategy written before the open.

The prefrontal cortex runs two jobs at once - holding the full option set, including the conditions that made the position worth taking, and inhibiting the pull to act on whatever is most salient on the screen. Both degrade as signalling efficiency falls, leaving a cheaper decision process. Each of the five failure modes we describe has a desk equivalent.

• Defaulting - holding, because holding requires no action. Inaction is the lowest-effort option on the screen and never neutral; it is a position chosen by not choosing.
• Simplifying - collapsing a conditional, hedged view into binary long-or-flat. The conditions were the analysis; dropping them feels like clarity.
• Snapping - acting on the tick rather than the thesis, justification assembled afterwards.
• Surrendering - deferring to whoever sounds most certain, because outsourcing a judgment costs less than forming it.
• Avoiding - the one the closing bell forbids, so it mutates into the other four.

The drift is not reliably toward more risk

It would be tidier if depletion made everyone bolder. It does not. One trader sits on a loser past the level where they said they would cut; another doubles size on a setup they would have skipped at 10am. Both are cheaper than deliberating.

Calibration is measurable against your written rules rather than against a feeling. None of this is a view on what to trade or how to size it; it is a statement about the machinery making the call.

What the evidence supports, and where it stops

The temptation is to quote a precise-sounding number. The good data is narrower than that.

The strongest published demonstration of a decision criterion drifting across a session comes from outside finance. Danziger, Levav and Avnaim-Pesso analysed more than a thousand parole rulings and found that favourable decisions declined markedly as a session progressed and recovered after food breaks. The direction is instructive: judges moved toward the status quo, which there meant denial - the option requiring least construction. The study has been debated, chiefly over whether case ordering explains part of the pattern, and that caveat is worth carrying.

Clinical medicine shows the same shape. Linder and colleagues (JAMA Internal Medicine, 2014) found clinicians prescribed antibiotics for respiratory complaints more often later in a clinic session - again a drift toward the lower-effort option.

Our own contribution is direct but bounded. We ran a randomised, placebo-controlled crossover trial in Phoenix, Arizona: 23 healthy adults, two 13-hour sessions of sustained competitive decision-making, ten matches per session, seven-day washout between arms, published in Frontiers in Nutrition in 2025. The Numin arm made 43% fewer decision errors and held baseline performance to the end of the observation window, with zero adverse events. The limits are worth stating: 23 participants is a small sample, it is a single trial, and the population was competitive gamers, not traders. The load profile is a fair proxy for a desk, but it is a proxy.

No public dataset breaks out trader P&L by hour of session with decisions attributed to it. Desks measure this internally and have no incentive to publish, so there is no honest percentage to quote for what the last hour costs. If you are shown one, ask where it came from.

Protecting the last hour of the session

The first defences are structural - they move decisions out of the depleted window.

• Write the rules while the system is fresh. Limits, exits and what would falsify the thesis cost far less to define before the open than to improvise at 3:30pm.
• Clear non-market decisions out of the afternoon. Admin, vendor calls and staffing draw on the same account as position sizing. The leadership version of the day is in first call to market close.
• Take a real break before the last hour. Recovery after breaks is one of the few effects visible in the parole data. Scrolling a feed is not a break.

Those tactics have a ceiling: you cannot move the close, and discipline does not restore signalling efficiency once it has fallen.

That ceiling is the gap Numin was built to address. It is a stimulant-free powdered drink taken 30 to 60 minutes after lunch. It takes about an hour to begin working and gives six or more hours of clarity, which places its window over the afternoon and the close; it is best avoided after 4pm. Rather than manipulating dopamine or norepinephrine the way stimulants do, it supports the brain's natural glutamate clearance system: the difference between overclocking a computer and preventing it from overheating. It prevents a decline from baseline rather than lifting performance above it - the accurate claim, and the useful one, since your rules were written at baseline. Five named ingredients, no proprietary blend, no caffeine and no sugar. The formulation detail is on the Numin science page, and a 20-pack is one month at $54 on subscription. A verified customer review from a trader puts it concisely: "Decisions before close are expensive if they're bad."

There is no dependency, tolerance or rebound; stopping returns you to ordinary afternoon decision fatigue. One boundary worth stating plainly: if your fog is present in the morning and after rest, it is not tracking decision load, and that warrants a clinician rather than a supplement. Nothing here is investment advice.

The close arrives at the same time every day, carrying the day's heaviest decisions. The variable you can change is the state of the system that meets it.