Technical analysis · Price reversal
Weekend Gap Reversal: Evidence, Entry Rules and Risk Controls
Weekend gap reversal trades against unusually large changes between Friday’s close and the following market opening. This analysis separates a published historical finding from the common assumption that every gap will close, then sets out a practical model for testing the idea.
What the research actually found
Dao, McGroarty and Urquhart examined 16 USD currency pairs using 2002–May 2014 data. Their trading test began in January 2007, with thresholds estimated from 2002–2006 and subsequently updated. They traded against extreme weekend gaps and exited at the end of the week. Assumed round-trip costs were 0.08% for major pairs and 0.30% for emerging pairs, with interest included. Weekend overreaction in spot FX (2016), sections 2–4.
Table 9 reports a 0.06% average weekly alpha for the major-currency portfolio under rolling estimation, significant at 1%. EUR/USD’s 0.04% was not statistically significant. Alpha is a regression estimate, not compound account growth. The emerging-currency group’s alpha was also not significant. Table 10 shows individual-pair drawdowns reaching approximately 24.1%. These mixed results do not establish that every pair or gap was profitable. Author manuscript, Tables 9–10.
This is historical evidence for investigating a conditional reversal strategy. It is not verification of current broker execution, and no new Trade-Forex backtest is presented here.
Understand what a weekend gap measures
Define a gap using one consistent instrument, price feed and weekend boundary. It is the change from the last valid pre-weekend quote to the first valid quote when that same feed reopens. A broker may label the reopening Sunday evening while another daily chart labels it Monday. Store UTC timestamps and broker session information rather than trusting the candle label.
For a pair quoted in its normal broker direction, calculate:
Weekend gap (%) = 100 × (opening midpoint / Friday closing midpoint − 1)
A jump from 1.1000 to 1.1055 is +0.50%, or 55 pips for EUR/USD. Selling the pair fades that upward gap. A downward gap produces a buy signal if it is sufficiently extreme. Using the normal quote direction consistently keeps the direction intuitive; do not mix inverted and non-inverted histories.
A gap can reflect a durable change in information. Elections, policy decisions and other weekend events can move the price to a new level. The Friday close is a reference point, not an obligation for the market to return there.
Define “large” without looking into the future
The educational model below uses the preceding 260 valid weekends for each pair. Before the current opening, calculate that pair’s 5th and 95th percentiles of historical percentage gaps. Use a documented quantile method, such as linear interpolation between adjacent sorted observations, and keep it unchanged.
Only past observations belong in that window. Including the current gap or later weeks in threshold estimation leaks information into the signal. A fixed 30-pip threshold has a different meaning across price levels and volatility conditions; this model therefore uses percentage quantiles.
If fewer than 260 valid weekends are available, skip the pair. Remove documented bad quotes according to a rule written in advance, retaining an audit log. A large genuine market move must not be deleted simply because it damages the strategy’s results.
Complete rules for a controlled demo test
These are our implementation choices. Delayed execution, protective stops, spread limits and exposure caps differ from the published test; its reported alpha cannot be assigned to this model.
| Decision | Fixed rule |
|---|---|
| Initial test instrument | EUR/USD, selected for a manageable execution exercise, not because the paper proved a significant edge in it |
| Upward trigger | Positive gap strictly above the prior 95th percentile: prepare a short |
| Downward trigger | Negative gap strictly below the prior 5th percentile: prepare a long |
| No trigger | Remain flat for the whole week |
| Entry time | First tradable quote at least 15 minutes after the broker’s weekly reopening |
| Spread screen | Skip the week if the spread at that entry time exceeds 2.0 pips; no later retry |
| Already reversed | Skip if the midpoint has reached or crossed Friday’s close before entry |
| Initial stop | Twice Friday’s completed daily Wilder ATR(14), measured from the actual entry |
| Exit | Stop-loss, or Friday at 16:45 America/New_York, whichever occurs first |
| Position handling | One entry per week, no averaging down, no re-entry after a stop |
There is no take-profit at the Friday reference price in this specification. Reaching that level after entry does not close the trade. The time exit lets a reversal extend beyond a full gap fill; it also allows an earlier unrealized gain to disappear. A gap-fill exit would be a different version and must be tested separately.
If the scheduled Friday exit is unavailable because of a known holiday session, use the broker’s published final tradable time minus 15 minutes, recorded before entry. For an unexpected interruption, use the first available executable quote and record the exception. Never report the scheduled price as an achieved fill when the market was unavailable.
A trade that works—and one that fails
Assume a fictional Friday close of 1.1000 and reopening midpoint of 1.1055. The upward gap is 0.50%. Suppose the previously calculated upper threshold is 0.35%, so the gap qualifies. At the model’s entry time, the midpoint remains above Friday’s close and the spread passes the screen.
Take an actual short fill at 1.1050. Friday’s completed ATR is 0.0040, or 40 pips, making the initial stop 1.1130, 80 pips above entry. An eventual Friday exit fill at 1.1010 produces 40 pips gross profit. The price has not even returned completely to the old Friday close; a full fill is not necessary for this short trade to profit.
If the ask instead reaches the protective stop and the closing fill is 1.1130, the trade loses 80 pips before commission and financing. Slippage can make the loss larger. A long trade reverses the directions: entry at ask, closing sale at bid, with the stop below the entry.
On a hypothetical $10,000 USD account, choose an illustrative risk budget of 0.25%, or $25. With $10 per pip per standard EUR/USD lot, an 80-pip stop and a $7 round-trip commission per lot give a size limit of 25 ÷ (80 × 10 + 7) = 0.03098 lots. Round down to 0.03 lots where supported.
At 0.03 lots, the favourable path earns $12 from the executed prices, less $0.21 commission and any financing. The stopped path loses $24 plus $0.21 commission and financing. The spread is already reflected when actual entry and exit fills are used; subtracting it again would double-count it. Reserve further room for slippage and swap when setting a final size.
All prices, thresholds and trade outcomes in this section are invented examples. They demonstrate mechanics, not a profitable track record.
Why execution can remove the apparent edge
Reopening quotes are especially important to this test: the measured gap and the tradable entry may be materially different. A chart midpoint is not a price at which both a purchase and a sale can occur. Reconstruct long entries at ask and short entries at bid, then use the opposite side for exits.
Track spreads during the 15-minute delay, missed trades, slippage and connection failures. Never fill a stop at its requested price if available quotes have jumped beyond it. Use bid/ask tick data to resolve the order of intraday events; daily OHLC alone cannot show whether a stop was hit before a later recovery.
Financing matters because the model can hold for most of a week. Apply the account’s actual long or short swap and its multiple-day rollover convention. A fixed historical cost assumption does not establish what this account would pay today.
One favourable example does not offset a poor payoff distribution. With an average 40-pip winner and 80-pip loser, the illustrative gross break-even win rate is 66.7%; positive costs raise it. The actual model has variable profits at its time exit, so estimate its real average winner and loser from all trades, not from these examples.
Validation, stopping conditions and common mistakes
Create a chronological development sample and an untouched final test sample. Record the thresholds, signal, spread screen, actual entry, stop, exit reason, financing and net P/L for every week, including weeks with no trade. Report calendar-time results as well as per-trade statistics so a rare signal does not appear busier than it is.
Assess net expectancy, maximum drawdown, time under water, number of trades and sensitivity to doubled costs. Check whether profitability survives excluding the single best year. The low frequency of extreme gaps means even several months of demo trading can supply very little evidence.
Use an illustrative 2% account drawdown suspension for the demo experiment: close any open position when equity is 2% below its running peak, then stop taking new trades until the experiment is reviewed. Report this overlay separately, and do not assume the threshold caps losses exactly. Do not extend the holding period or add size merely because a gap has not closed.
Do not claim replication unless you also reproduce the paper’s data conventions, portfolio construction, financing and original timing. Do not select the best historical currency after seeing its result and present that choice as independent validation. Reject this model if out-of-sample net results fail, even though some historical research results were positive.