Sleep, fatigue and stress in trading: state as an input, not an explanation
Sleep, fatigue and stress can affect attention and task performance, but they do not automatically explain the result of a particular trade. You can profit after breaching a plan and lose despite following it correctly. This article therefore separates subjective state, process quality and financial outcome. It examines what sleep-deprivation research actually shows, how to record your state before a session, and how to compare it with breaches without overlooking time of day or market conditions. It then translates those observations into breaks and conditions for limiting new risk. It also marks the point at which another trading rule cannot replace a discussion with a clinician or another appropriate qualified professional.

Check whether the state record existed before the outcome
A diary can contain a state label without containing a usable pre-session observation. The timing matters. A label added after seeing the result answers a different question from a label recorded before the session. Begin by checking that distinction, before comparing groups or explaining a difficult day.
Consider a hypothetical review of 30 scheduled sessions. In 24 sessions the state form was saved before the prewritten deadline; in 6 it was late or missing. Timely coverage is 24/30=80%. Keep the other records visible as late or unclassified. Do not reconstruct their earlier state from whether the session later made or lost money.
The example’s labels are personal descriptions used for this log, not medical categories. The comparison below examines completion of named process fields. It is not a cognitive test, a diagnosis or an assessment that a person is safe to trade. A session without a trade still belongs in the scheduled-session register.
Define the unit before calculating an error rate
Among the timely records, 6 sessions carry the trader’s pre-session “more fatigued” label and 18 carry the other label. Every session has the same 5 required preparation fields, whether or not an order is eventually sent. An omission means a required field was not completed by its deadline under the unchanged logging rule.
| Pre-session label | Sessions | Required fields | Omitted fields |
|---|---|---|---|
| More fatigued | 6 | 30 | 3 |
| Other label | 18 | 90 | 6 |
| All timely records | 24 | 120 | 9 |
The first group has 3/30=10% omitted fields. The other has 6/90≈6.67%. Saying that the second group had more omissions because its raw count is larger leaves out its larger number of opportunities. Across the timely records, 9/120=7.5% of required fields were omitted.
These are rates per field, not percentages of sessions with any breach. Several omissions may occur within one session. To calculate a session-level rate, retain which sessions contained them. The table also does not establish that the observations are statistically independent.
Keep three different measurements in separate columns
The state label, completion of preparation fields and financial result are different variables. A completed form does not establish a good trade, and a profitable trade does not prove every field was completed. Record the result separately rather than using it to repair a missing earlier observation.
Van Dongen and colleagues’ laboratory experiment assigned time-in-bed conditions and repeatedly measured neurobehavioural performance and subjective sleepiness. The distinction between an assigned condition, a self-report and a measured task outcome matters. That experiment did not test this diary, establish a trading-profit effect or provide a personal entry-permission threshold.
Our table contains invented observational records. It describes an association under a chosen process definition. It cannot tell whether fatigue caused the omissions. Session time, workload, task difficulty, market conditions and the way the label was used may also matter. The 6 late or missing state forms remain another source of uncertainty, rather than evidence of no difficulty.
Choose the next review from the limitation you actually found
The first repair in this example is data quality: retain the deadline, the original timestamp and an explicit missing-state category. The next is measurement consistency: keep the same required fields and preserve any authorised revision as a new version. Changing what counts as an omission halfway through the review changes the comparison.
If you examine an association further, keep the schedule and relevant context with every observation. Do not convert the difference between these field rates into a probability of losing money or a claim that a break will cure it. A change to the trading procedure needs its own rationale and later review; it is not a clinical intervention.
The detailed lesson below examines sleep and stress research, confounding and the limits of a trading diary. Use the introductory audit to distinguish what was recorded on time, what was actually measured and what remains unknown. Those questions make the review more informative without pretending that a small personal log can settle a health or causal question.
Three exercises with worked answers
The exercise data are hypothetical. Choose an answer, then check the calculation.
From a clean observation record to a careful reading of the research
The full lesson retains its original illustrations, counterexamples, research limitations and confounding calculation. Use it to distinguish laboratory findings, a personal process audit and questions that require a separate professional assessment.
1. State before a decision is data, not an explanation
The explanation “I was tired and that is why I did it”, recorded only after a loss, is insufficient to establish what caused the decision. It already uses knowledge of the outcome and can easily be fitted to the event instead of considering alternatives. A state recorded before the session provides a better reference: a column by which trades can later be grouped and execution compared. The entry still establishes no cause on its own, but lets you check whether observed events vary alongside state.
Here, sleep, fatigue and stress are treated as possible influences on execution rather than ready-made explanations for every decision. Research can identify associations or effects in specific tasks, but a single order history does not establish a cause. In the journal, these are input variables alongside time of day, instrument and earlier decision counts. A pre-session record is most useful because it lets you examine hypotheses without tailoring your assessment of state to the subsequent outcome.
Reduced executive performance – how to recognise it in the record
We define degraded process execution so that it can be checked in order history and the journal. It means departures from the same trader’s previously recorded plan. This is an operational process measure rather than a diagnosis of cognitive impairment or an executive-function test score. Observable examples include:
- an order placed without the journal entry that was supposed to precede it – the order timestamp is earlier than the note timestamp, or there is no note at all;
- a skipped field on the entry-conditions list while the rest of the pattern is preserved;
- a position size departing from the calculator output, with the rule unchanged;
- a protective order set after entry confirmation instead of before it;
- going back to the chart and reversing a decision not to enter, even though the condition is still not met;
- an extended time between a condition appearing and the reaction to it, with the event type unchanged.
Every one of these points is decidable from a history export and the journal. None requires you to assess your own state, and none assumes that fatigue was the cause. That is the purpose of such a definition: to describe what happened, before you start settling why.
It is worth separating two things straight away that popular texts merge into one. The feeling of fatigue and degraded execution are two different variables and need not occur together. You can run a flawless session while feeling terrible, and you can wreck the whole process while feeling great. Self-assessment is a useful subjective measurement – comparable with your own earlier entries, incomparable with anyone else's – but it is not a measurement of performance. The entry "state 3 out of 5" written before the session is data. The same entry made in the evening, once the day's outcome is known, is a story about the outcome.
2. A counterexample from both sides: the rested trader breaks the process too, the tired one executes it too
The examples below address a narrower claim: good sleep does not guarantee adherence to a plan, and short sleep does not make every decision erroneous. Two individual cases neither refute a statistical effect of sleep deprivation nor establish safe trading conditions.
Illustrative example, first variant. A full night, a morning session, the second day of a quiet market. The pre-entry record lists four setup conditions; three are met. The trader enters, adding to the note that the fourth condition "does not apply in these circumstances". Position size matches the calculator, the protective order is in place, the journal entry precedes the order. The trade ends in profit. State: impeccable. Outcome: positive. Process: broken at one specific point – this is what the example shows. Good sleep did not guarantee adherence to the plan. The record reveals a change in the criterion after looking at the chart, but does not establish its psychological cause.
Second variant. Four hours of sleep, an evening session. Setup conditions met in full, size from the calculator, protective order before confirmation, journal entry timestamped earlier than the order. The trade ends in a loss. Process: complete. Outcome: negative. In line with what article 10.4 calls separating process from outcome, this loss is not evidence of an error, and certainly not evidence that lack of sleep was to blame.
The two variants show that neither outcome nor adherence to a plan can be equated with physical state. They do not establish that sleep is irrelevant. State is one potentially important variable to record before a decision and then compare with execution across a larger series. Subjective state can also be recorded beforehand rather than reconstructed after a loss. Checklist adherence is not a complete measure of cognitive performance or safety.
3. What the research shows – and what is not in it
The literature on sleep deprivation is extensive and partly well replicated. The problem lies elsewhere: between what was measured in the laboratory and the sentence "your performance will drop" there are several rungs that nobody has climbed. The table below gives, for each item, the population, the nature of the task and the effect size – and separately, what does not follow from it.
| What was measured | Source, population, sample | Direction and size of the effect | What follows for a trader – and what does not |
|---|---|---|---|
| Short-term total sleep deprivation across six categories of cognitive function | Lim and Dinges, 2010; meta-analysis of 70 articles and 147 tests; adult volunteers, laboratory conditions, no financial stakes[1] | Largest effect for lapses of attention in simple tasks: g ≈ 0.78, confidence interval [0.60; 0.96]. Smallest for reasoning accuracy: g ≈ 0.13, interval [−0.02; 0.27], given as the magnitude of impairment – it crosses zero, which is why the effect is not significant | Follows: in these laboratory tasks, attention lapses were particularly vulnerable to total sleep deprivation. Does not follow: that a trader retains full setup-analysis capacity or that a prepared list confers resistance. A non-significant reasoning estimate does not prove no effect, and trading checklists were not tested. |
| Sleep deprivation and decision making | Harrison and Horne, 2000; literature review; laboratory studies, mixed populations[2] | Rule-based, logical and convergent tasks under short deprivation often remained unchanged, presumably thanks to compensatory effort. Impairment concerned unexpected situations, revising a plan, acting under distraction, and communication | Follows: tasks differ in susceptibility, and revising plans or responding to the unexpected deserves attention. Does not follow: proven resistance of a trader’s checklist. Recording conditions beforehand, as in 10.7, is a design inference rather than a safety test after sleep loss. |
| Sleep deprivation and higher-order executive functions | Killgore, 2010; review; laboratory studies[3] | Consistency of results is low: for perception, memory and higher-order executive functions the literature remains equivocal | Follows: there is no single "hours of sleep versus performance" curve you could read off for yourself. Does not follow: that there is no effect – what follows is that it is more poorly defined than popular write-ups suggest |
| Extended wakefulness and a psychomotor task | Dawson and Reid, 1997; 40 participants, crossover design; hand-eye tracking task; laboratory, no financial stakes[4] | Impairment after 17 hours of wakefulness matched the impairment at a blood alcohol concentration of 0.05%, and after 24 hours – about 0.10% | Follows: an order of magnitude for one narrow function, measured by one task. Does not follow: that a trading decision after a sleepless night is "like being drunk" – the equivalence was established for hand-eye coordination, not for assessing entry conditions |
| Cortisol and testosterone versus results on a trading floor | Coates and Herbert, 2008, PNAS; 17 traders, male, one London trading floor, 8 consecutive working days, saliva samples[5] | Cortisol rose with the variance of the trader's own result and with market volatility. No relationship between cortisol level and P&L level was found (t = 0.12; P = 0.9; n = 14) | Follows: the marker tracked market conditions. Does not follow: that stress worsens financial results – this paper did not show that, although it is cited for exactly that purpose |
The chart shows different effects across tasks after short-term total sleep deprivation, rather than after any short night. Impairment in attention lapses was substantial. The estimate for reasoning accuracy was smaller and not statistically significant, but its confidence interval does not establish a complete absence of an effect. Other categories, including working memory and complex attention, also showed impairment. Traders’ checklists were not tested. The practical inference is cautious: vigilance deserves attention, and a prepared condition list may limit improvisation, but it does not confer resistance to sleep loss or guarantee correct analysis.
Sleep is not a single number
The question "how many hours of sleep does a trader need" is badly posed for two reasons. First: it is a medical question, and health norms are not the subject of an investing course. Second, and more important here: sleep has at least five dimensions, and each of the studies cited measured a different set of them.
- duration – the only dimension that usually makes it into the journal, and one of the dimensions you can record without equipment;
- continuity – a night broken three times by checking an open position is not the same night as uninterrupted sleep of the same length;
- timing – the same number of hours slept in a different window of the day is a different variable;
- regularity – the spread of bedtimes across the days of the week;
- one-off versus cumulative deficit – one short night and the fifth short night in a row are not the same event.
The “hours of sleep” field captures duration alone and omits the other dimensions; it does not measure them together. It is useful, but supports a narrower conclusion than a complete sleep assessment. You can additionally record timing, continuity or subjective quality if you can maintain those fields. A wristband is not required, although device measurements may have their own uses and limitations.
4. Decision fatigue and the hungry judge: a contested effect
The most frequently repeated "proof" that a series of decisions degrades subsequent decisions is a study of Israeli parole boards. Danziger, Levav and Avnaim-Pesso described in 2011 a pattern in which the share of favourable rulings falls from about 65% at the start of a session block to near zero before the break, and then returns to its starting level[6]. The story is vivid, sells well, and regularly serves in trading texts to justify specific time thresholds.
It has, however, two serious methodological criticisms, and neither has been closed.
Weinshall-Margel and Shapard showed that the conclusion depends on the assumption that cases were ordered randomly – and the order was not random. Cases from a single prison were heard in a block before the break, and after the break they started with a different prison; on top of that came the presence of legal counsel, correlated with position in the queue[7]. A few years later Glöckner showed by simulation that the way the sessions are scheduled is by itself enough to generate the observed pattern: favourable rulings take longer, so towards the end of a block there is simply not enough time left for them[8]. The authors of the original paper responded to the first criticism by recoding the data and maintained their conclusion. The dispute continues.
Separately, it is worth splitting two things that constantly get mixed up in this context. Fatigue may build up over time and repeated actions, but its course depends on the task and conditions. Time pressure is a property of the situation and can occur in the first minute of a session, after the best night of your life – all it takes is a data release in ten minutes and an unfinished list of conditions. The countermeasures differ: the first situation calls for limiting the number of decisions, the second – for a decision taken earlier, in the form of a pending order or a written condition. Confusing them leads to a trader sleeping properly and still taking the same entries without a full set of premises.
5. Before you blame a worse day on fatigue – a calculation over a series of sessions
The following is an illustrative dataset rather than actual observations: imagine forty sessions with a "sleep" field filled in before the session and with the start time recorded. As the measure of process you take the number of procedure breaches – events from the list in section one – and not the financial result, because the result depends on the market while a breach depends on you. You split the set by sleep and get this:
| Split by sleep | Sessions | Breaches | Breaches per session |
|---|---|---|---|
| Typical sleep | 22 | 15 | 0.68 |
| Short sleep | 18 | 21 | 1.17 |
| Total | 40 | 36 | 0.90 |
The difference looks convincing: 21 breaches over 18 sessions against 15 over 22 is 1.71 times more breaches after short sleep – a ratio computed from the counts, not from the rounded figures in the last column. This is where most analyses end and the resolution to get more sleep begins. So let us add one column we already have in the record anyway – the session start time.
| Session time | Sleep | Sessions | Breaches | Breaches per session |
|---|---|---|---|---|
| Morning session | typical | 18 | 9 | 0.50 |
| short | 6 | 3 | 0.50 | |
| Evening session | typical | 4 | 6 | 1.50 |
| short | 12 | 18 | 1.50 |
Within each time of day the difference between short and typical sleep is zero. The whole difference in the first table came from the fact that sessions after short sleep were mostly evening sessions: 12 out of 18, that is 66.7%, against 4 out of 22, that is 18.2%, in the other group. The variable associated with the difference in this dataset is session time – 0.50 in the morning against 1.50 in the evening – and sleep merely co-occurs with it.
How much data you need to compute this at all
Here a second problem appears, one usually passed over in guides about journalling. Forty sessions split by one variable give two groups of twenty observations on average. A split by two variables – four cells of ten on average, with the smallest in our table holding four sessions. Adding the instrument gives eight cells of five on average, and adding the volatility regime – sixteen cells of two and a half.
In this example, each additional binary split doubles the number of cells; that is not a property of every method for controlling variables. Potential confounders include instrument, time of day, volatility, earlier decisions and outcomes, and position size. More small groups mean fewer observations in each. Calculate the pattern and treat it as a hypothesis to check with further observations rather than a conclusion after forty sessions. The sample needed depends on variability, the size of the difference and the method. Article 10.4 discusses limits to inference; four observations in a cell yield a highly uncertain estimate.
The tools for such an analysis are usually within reach: external analytics services can group trades by opening hour with no work on your part. What they will not do is add a column with the state recorded before the session – because that data point simply is not in the order history. A human has to enter it before opening the platform.
6. Four different things under one word, "stress"
The word "stress" in trading texts works as a skeleton key: it fits everything, so it opens nothing. Underneath it hide at least four phenomena with different symptoms and different countermeasures.
| What it actually is | What it looks like in behaviour at the platform | What to do about it – and where it is described |
|---|---|---|
| Time pressure – a property of the situation, not of the state | Shortening the list of conditions because "it is about to go"; an order placed before the check is finished | A decision taken earlier: a pending order or a condition written down before the event. The construction is in 10.7 |
| Arousal – independent of whether the outcome was good or bad | Shorter intervals to the next decision, more orders per unit of time, position size outside the rule – including after a winning streak | An interruption condition written down before the episode. The whole procedure is in 10.9, the mechanics of escalation in 10.5 |
| Cognitive load – the number of things held in your head at once | A skipped list field, the wrong instrument, open positions the trader forgets are there | Shortening the list and reducing the number of simultaneous exposures; designing the checklist is in 10.7 |
| Subjective stress – what the trader writes in the self-assessment field | It has no constant behavioural manifestation, and that is precisely its diagnostic feature: it is sometimes high in a session executed flawlessly | Record it before the session as data comparable with your own earlier entries; do not treat it as a measurement of performance or as a basis for deciding on an entry |
The distinction between acute and chronic stress concerns the duration and course of strain, rather than a threshold in order history. An acute reaction may follow a particular event, while prolonged strain can persist or recur over an extended period. Changes in decision intervals or a persistently elevated breach rate are execution observations, not diagnostic criteria for chronic stress. Record perceived stress separately from the process. If symptoms persist or interfere with daily functioning, the journal cannot replace a discussion with a qualified professional.
7. The pre-session state record: three fields, not a questionnaire
Everything I have written so far leads to one operation: before you open the platform, you write down a few values that cannot honestly be reconstructed after the session. Not in order to grade yourself. In order to have a column to split by forty sessions from now.
The minimal version has three fields and fits on one line:
- Hours of sleep – a number, entered before the session. One number, in full awareness that it glues together the five dimensions described in section three;
- Start time and the number of this session that day – the most important column of the three, because it is the one that most often turns out to be the real explanation;
- Number of decisions taken earlier that day – counting those that ended in deciding not to enter, because those were decisions too.
An optional fourth field records self-assessed state on a five-point scale. Interpret the difference between 3 and 4 through the meaning of your earlier entries; do not treat it as an independently established psychological quantity. A more extensive form may be harder to maintain. Choose fields whose purpose you can explain and check their completeness. The full journal structure, time layers and breach codes are described in article 10.8. Here we add three core columns and an optional self-assessment.
What this record is not for. It is not a clinical test, does not classify you and cannot confirm fitness from one number. Wristband or watch readings can provide additional data if they retain pre-decision measurements, but accuracy and interpretation require care. This procedure supplies no heart-rate, HRV or sleep-quality score thresholds for safe trading: the cited sources do not validate them. The record supports an execution review rather than replacing a health assessment.
8. Breaks: limiting the number of decisions, not resetting the body
A break can be specified in the session plan: when it begins, what it involves and when it ends. Here, we treat it neither as a biological “reset” nor as a guarantee of regaining your morning state. The cited studies do not establish how long a trader would need to “reset the nervous system”. We therefore assess a planned break through records of its execution and subsequent events rather than attributing an unmeasured mechanism to it.
A break followed as planned excludes new entries during that interval and lets you change the context of the next decision. It does not guarantee fewer orders over the whole session: they may be made up afterwards. Check both break adherence and total decision counts, intervals and subsequent breaches. Existing positions and protective orders must continue to be managed under the previously established plan. This describes a procedural mechanism, not a promise of a particular biological effect.

The second thing that must be said plainly: there are no studies measuring the effect of breaks on the quality of trading decisions in retail traders. Everything below is craft practice and should be treated as such. That does not make it worthless – it means the authority behind these sentences is the construction of the procedure, not science.
Why a break must be procedural rather than discretionary
The best argument for writing breaks into the plan does not appeal to physiology but to who makes the decision about the break: a break decided during the session is decided by that session. The whole construction built on this argument is set out in article 10.9; here one thing follows from it: the break has to be written into the plan for the day and have an external signal – a timer, an alarm, anything that does not require your consent at the moment it falls due.
Sleep research indicates that sustained attention is particularly vulnerable after total sleep deprivation[1]. That is a reason to examine long periods of screen monitoring, but not proof that these are every trader’s riskiest periods. A break can interrupt a monotonous activity; the cited meta-analysis does not establish its effect on entry quality. A prepared condition list can organise work, but reasoning-test results do not show that it protects against the effects of sleep loss.
Parameters this article will not set for you
How long a break should last and how often it should fall – I do not know, and nobody has measured it for retail trading. Any number of minutes given here would be invented and would, incidentally, replace a value fitted to your session with a value fitted to nobody's. There is one design principle that can be defended: the interval must be fixed before the session and be the same regardless of what the market is doing. A break cancelled "because there is a big move today" is not a break, it is a declaration.
You enter the interval yourself and treat it as a parameter to be checked after a few dozen sessions – by exactly the method described in section five, with a split by time of day, so that you do not attribute to breaks something the hour did.
Finally, the boundary this section does not cross. A procedural break written into the plan for the day and stopping trading after an episode are two different things: the first falls due always and has no trigger condition, the second is triggered by an event and has a return condition. The second belongs entirely to article 10.9 and I will not repeat it here.
9. The rule "I do not trade when condition X holds"
The advice “get enough sleep” does not specify what to do or how to check execution for a particular decision — for example, at five on Friday afternoon. A work-organisation rule answers four questions: what triggers it, what you do, when it ends and what record you leave. Separately, define how you will recognise that it needs changing. Missing elements make its operation harder to assess. This is the form used throughout the section; its construction is explained in article 10.7.
| Element | What it answers | Example of a filled-in field – illustrative variant |
|---|---|---|
| Trigger condition | What exactly must happen for the rule to switch on – checkable before the session, without assessing your own state | The hours-of-sleep field entered before the session is below a previously recorded working threshold N, or the plan excludes a second session that day. N illustrates a work rule, not a health norm derived from forty sessions. |
| Action | What specifically happens – an action, not an attitude | I do not open new positions. Managing already open ones continues unchanged, in line with the plan made before entry |
| Termination criterion | When the rule stops applying – without this it cannot be executed | From the first session of the next day, with the sleep field at least N and the remaining recorded conditions met. Feeling better alone does not replace those criteria. |
| Audit trail | What is left in the record so that it can be checked whether the rule worked | One line: date, field value, the blocking decision, the number of opening orders placed in spite of it |
| Failure criterion (a requirement separate from the form) |
How you will recognise that the rule is to be rewritten rather than repeated | It has not triggered once in several dozen sessions, or it triggers every day, or it triggers and is worked around every single time |
Threshold N in the table is an illustrative work-organisation parameter rather than a health norm or confirmation of safety once it is exceeded. Do not determine your sleep needs from P&L or a few breach-free sessions. Financial risk and exposure limits belong to section 9; they do not establish sleep norms. Your record can help assess a working rule’s execution, but cannot replace health guidance or assessment of persistent symptoms.
The rule also has a cost: it restricts participation and may exclude situations in which all entry conditions are met. Record those potential opportunities, but do not assign them a certain result or assume that the entire skipped session would have been flawless. The blocked-opportunity count measures reduced activity rather than proving that the threshold takes away more than it protects. A review considers the rule’s purpose, execution, comparability of the data and health limitations. Article 10.7 discusses excessive rigidity, but a missed opportunity alone does not justify bypassing a rule during the session.
Chronotype and session time
Fitting your trading hours to the time of day in which you usually execute the process better is a sensible piece of craft practice – and I label it as such, because there is no evidence for it in retail traders. It does, however, have one advantage over most advice in this area: it can be checked on your own data, by exactly the same split we made in section five.
The caveat is the same as before and returns for the last time: time of day brings with it the character of the session, liquidity, volatility and the type of events you react to. When the split shows that you execute the process worse in the evening, you do not yet know whether that is your time of day or the properties of the market at that hour. What you do know is something more practical: that at that hour you execute the process worse – and that is enough to formulate a rule, though not enough to explain the cause. A rule can be correct even when its justification is unknown; demanding an explanation of yourself before acting is an unnecessary cost here.
10. Where this article ends
Everything above concerns one thing: the record of decisions and what can be read from it. There is, however, an area in which this way of working stops being the right tool, and insisting on it becomes a way of postponing something else.
It is also worth naming the opposite way of crossing this boundary, because in trading texts it is more common than diagnosing. It consists in treating everything above as a problem to be solved with yet another rule – one more threshold, one more block, one more version of the journal. If difficulties and rule circumvention recur despite changes, reviewing the procedure alone may be insufficient. There is no universal threshold of three attempts; professional help can be sought earlier.
What is not in this article and will not be: an edge arising from being well rested. If a strategy has no positive justification, no amount of execution will fix that, rest alone does not change its expected value after costs. What can be built here is more modest and quite concrete: a record of state before the decision in which, after forty sessions, you can check whether time of day and market conditions do not explain what you are attributing to fatigue. The map of the whole section and the order of the articles is kept on the section page, and all the thresholds this article deliberately does not set – in section 9.
FAQ – sleep, fatigue and stress in trading
How many hours of sleep do I need in order to trade?
After a sleepless night, can I sit down at the platform at all?
How am I supposed to know that a worse day was fatigue and not the market?
What about coffee, supplements and things that improve concentration?
Is it worth measuring sleep quality with a wristband or a watch?
Should I match my session to my chronotype?
Does "decision fatigue" simply exist?
Sources and bibliography
- Lim J., Dinges D.F., "A Meta-Analysis of the Impact of Short-Term Sleep Deprivation on Cognitive Variables", Psychological Bulletin, 136(3), 2010, pp. 375-389. A meta-analysis of 70 articles covering 147 cognitive tests in six categories; adult volunteers, laboratory conditions, no financial stakes. The largest impairment concerned lapses of attention in simple tasks (g ≈ 0.78; confidence interval [0.60; 0.96]), the smallest – reasoning accuracy (g ≈ 0.13; [−0.02; 0.27]). The intervals are given as the magnitude of impairment, with the sign reversed relative to the source notation (g = −0.125; 95% CI [−0.27; 0.02]); the interval for reasoning accuracy includes zero, which is why the effect is not significant. Transferring this to trading decisions is an extrapolation and was not the subject of the study. Source.
- Harrison Y., Horne J.A., "The Impact of Sleep Deprivation on Decision Making: A Review", Journal of Experimental Psychology: Applied, 6(3), 2000, pp. 236-249. A literature review. Rule-based, convergent and logical tasks under short sleep deprivation often remained unchanged; impairment concerned responding to the unexpected, revising plans, acting under distraction and communication. The authors note that few studies concern realistic high-level decisions – which is also a limitation when relating the results to trading. Source.
- Killgore W.D.S., "Effects of sleep deprivation on cognition", Progress in Brain Research, 185, 2010, pp. 105-129. A review covering vigilance, perception, emotion, learning, memory and executive functions. Cited in this article for one conclusion: for higher cognitive functions the literature remains equivocal, so there is no single "hours of sleep versus performance" relationship you could read off for yourself. Source.
- Dawson D., Reid K., "Fatigue, alcohol and performance impairment", Nature, 388(6639), 1997, p. 235. 40 participants, crossover design: wakefulness up to 28 hours in one condition, alcohol consumption up to a concentration of 0.10% in the other. Dependent variable: score on a hand-eye tracking task. The authors give two reference points: impairment after 17 hours of wakefulness matched a blood alcohol concentration of 0.05%, and after 24 hours – about 0.10%. This establishes no universal conversion for other waking durations or other tasks. The equivalence was established for one psychomotor task in a laboratory, without financial consequences; it does not apply to assessing conditions for entering a trade. Source.
- Coates J.M., Herbert J., "Endogenous steroids and financial risk taking on a London trading floor", Proceedings of the National Academy of Sciences, 105(16), 2008, pp. 6167-6172. 17 traders, male, one London trading floor, 8 consecutive working days, steroid measurement in saliva. Morning testosterone was associated with the afternoon result (t = 3.03; P = 0.008; n = 17). Cortisol correlated with the variance of the trader's result (r² = 0.48; P = 0.004; n = 17) and with the market's implied volatility (r² = 0.86; n = 11). No relationship between cortisol level and P&L level was found (t = 0.12; P = 0.9; n = 14). A correlational study on a very small sample from one team; it does not settle the direction of the relationship and does not show that stress worsens financial results. Source.
- Danziger S., Levav J., Avnaim-Pesso L., "Extraneous factors in judicial decisions", Proceedings of the National Academy of Sciences, 108(17), 2011, pp. 6889-6892. An analysis of rulings by Israeli parole boards; the share of favourable rulings fell within a block of sessions and returned to its starting level after a break. Cited exclusively together with items [7] and [8]; it is not the basis for any time threshold in this article. Source.
- Weinshall-Margel K., Shapard J., "Overlooked factors in the analysis of parole decisions", Proceedings of the National Academy of Sciences, 108(42), 2011, p. E833. A methodological criticism of item [6]: the conclusion requires a random order of cases, and the order was not random – cases from one prison were heard in a block, after the break they started with another, and on top of this came the presence of legal counsel, correlated with the ordering. The authors of the original paper responded to this criticism by recoding the data and maintained their conclusion; the dispute remains unresolved. Source.
- Glöckner A., "The irrational hungry judge effect revisited: Simulations reveal that the magnitude of the effect is overestimated", Judgment and Decision Making, 11(6), 2016, pp. 601-610. A simulation analysis showing that the way sessions are scheduled alone – favourable rulings require more time – is enough to generate the pattern described in item [6], without appealing to decision depletion. Source.
- World Health Organization, International Classification of Diseases ICD-11, version in force since 2022: chapter 07 (sleep-wake disorders) and category 6C50 (gambling disorder). Cited exclusively as an indication that the boundary between education about decision making and clinical diagnosis exists and roughly where it runs. It is not a tool for self-diagnosis and is not used in this article to assess the reader. Source.