What Are Support and Resistance Levels, and Do They Work?
Support and resistance are price areas where traders expect a move to stall, and a Federal Reserve Bank of New York study measured how often intraday currency trends actually stopped at published ones.

Support and resistance levels are price areas where traders expect a move to stall: support below the market, resistance above it. The CFTC's futures glossary defines support as "a price area where new buying is likely to come in and stem any decline", and resistance as its mirror image, where new selling turns up to blunt a rise. Both entries say area rather than price, which rules out the one-pixel line.
Whether the levels do anything is a separate question from what they are, and it has been measured. Carol Osler, then a senior economist at the New York Fed, collected the levels six firms — commercial and investment banks, and news services — published daily for three currency pairs, and tested them against thousands of randomly generated levels. Her results, and the order-book study she wrote afterwards, are where the evidence below comes from, including the parts of the convention that failed the test.
What are support and resistance levels?
A support level is a price area below the market where buying is expected to halt a decline; a resistance level is the same idea above the market, where selling is expected to cap a rise. The CFTC's glossary gives both in those terms and cross-references each to the other.
The trading manuals Osler surveyed agree on the substance and add one qualification worth keeping: the halt is temporary. The definitions she quotes from Murphy, Pring and Arnold describe demand or supply concentrated enough to stop a move at least for a while, which is a much weaker claim than a turning point. Her review found little disagreement across seven manuals published between 1978 and 1997, and dates much of the standard indicator set to a 1930 book by Shabacker, written from observations of US stock prices.
Two further details come straight from her data. The firms in her sample usually published several support and several resistance levels a day, so "the level" is rarely a single number anybody agrees on. Support and resistance were also the only signal all six of those firms provided, with some providing nothing else — which tells you how central the convention is to how price gets discussed, independent of whether it predicts anything.
The CFTC glossary is blunter about the whole category. Its entry for technical analysis describes forecasting from patterns of price, rates of change, volume and open interest, setting fundamental factors aside, and ties the method's consistency to prices not following a random walk.
How do traders decide where a level sits?
Levels come from a short list of inputs, and Osler's review of the manuals collects them: recent highs and lows, round numbers, fractions of a prior move, trendlines, and private knowledge of where orders sit. Taking them roughly from the most mechanical to the most judgemental:
- Prior turning points. A previous reaction low marks support and a previous peak marks resistance. Osler's review calls looking at recent minima and maxima the simplest of the visual approaches.
- Round numbers. Whole figures get used as price objectives, so they end up acting as levels through sheer habit. Osler's own data bears this out: more than 70% of the published levels in her sample ended in 0 and 96% ended in 0 or 5, against the 10% and 20% you would expect from random choice.
- Fractions of a prior move. The "50 percent rule" holds that the first correction after a large move often gives back about half of it; Fibonacci retracements of 38.2% and 61.8% are used the same way.
- Trendlines and channels. Joining recent peaks with one line and recent troughs with another produces sloping levels rather than flat ones, which one of the manuals she cites calls a support and resistance zone.
- Knowledge of where orders sit. An analyst who hears that exporters are selling at a particular rate can publish that rate as resistance. This is information about order flow rather than about the chart.
Because the inputs differ, the outputs differ. Across the fifteen pairs of firms in Osler's sample, only about 30% of possible matches between two firms' levels were realised when a match meant agreeing within five points, and about 18% at a two-point tolerance. A pivot point is a different kind of object: its levels come out of a documented formula, so two traders running it over the same period draw the same lines.
What does a charting platform actually draw?
A charting platform draws the extremes and the session windows its inputs tell it to draw, and those inputs are yours to set. Sierra Chart's Swing High and Low study spells the arithmetic out: with a Length input of n, a bar is a swing high when its high exceeds the highest high of the n bars before it and the n bars after it, and a swing low is the mirror image using lows.
The consequence of that definition is easy to miss. Because the test looks forward as well as back, the study only resolves a bar once n more bars exist, and the arrow is placed retrospectively on the bar n back. A swing high that you are reading as resistance was confirmed as a swing n bars after it formed.
Session levels work differently. Sierra Chart's High/Low for Time Period study draws a line each day at the highest and at the lowest price over the window between a Start Time and an End Time you set, with no reference to the session times configured for the chart. The Extended version holds the same lines unchanged from a stop time until a separate line stop time, and reads a stop time earlier than the start time as the next trading day. The page also warns that a window falling inside a bar returns that bar's high and low, so fifteen minutes specified on a one-hour chart gives the one-hour extreme.
Quantower's ZigZag page connects its own output to the convention: it says the indicator's highs and lows, taken together, can serve as a support or resistance level, and that traders who use it that way group the nearest swings into one price. The deviation setting that decides which swings exist at all defaults to 5%, so a move smaller than that is not drawn, and changing the number changes the levels.
Do support and resistance levels actually work?
Published levels interrupted intraday currency trends more often than randomly chosen levels did, in the test Osler's article sets out. She collected the levels six firms published daily for the German mark, Japanese yen and British pound against the dollar from January 1996 to March 1998 — roughly 23,700, 22,800 and 17,700 values respectively — and compared what rates did at those levels with what they did at artificial ones.
The definitions she used are worth stating, because they decide what the result means. A hit was the bid or ask coming within 0.01% of a level. A bounce was the rate sitting back on the original side of that level fifteen minutes later. The bounce frequency is bounces divided by hits. For each day she generated 10,000 sets of forty artificial levels — twenty supports and twenty resistances — drawn at random inside a range taken from that month's largest move away from the open, and then counted how many of the sample's 28 months the published levels beat the artificial average in.
| Exchange rate | Bounce frequency at random levels | Published levels |
|---|---|---|
| Dollar-mark | 54.9% | +4.2 pp above random |
| Dollar-yen | 57.3% | +5.6 pp above random |
| Dollar-pound | 56.3% | +4.0 pp above random |
| All three combined | 56.2% | 60.8% |
Every one of the sixteen firm-currency pairs came out above its artificial benchmark, and the margin was statistically significant at the 5% level in all but three of them. Firms varied: the best and worst differed by about 4.0 percentage points on average, with one firm's yen levels running 9.2 points above random. The signal also aged slowly — five trading days after publication the levels were still ahead of random for every firm and currency, still significant in nine of those sixteen pairs, with bounce frequencies about 1.7 points below their publication-day figures.
What does the bounce test not prove?
It does not show that trading the levels makes money. Osler did not examine profitability; her conclusion names it as future work, noting that profitable bounces would give traders an incentive to arbitrage them away. Three more limits matter as much as the headline, and scope sits outside the test altogether.
The baseline is not a coin flip. Random levels bounced 56.2% of the time, where a random walk would have given about 50%. Her endnote names two reasons: a negative first-order autocorrelation in the rate changes, which she calls fairly strong, and the hit-then-stay construction of the test itself. The published margin sits on top of that baseline, at about four to six percentage points.
The firms' own strength ratings did not hold up. Three of the six graded their levels for strength on a three-point scale. In many months the strongest grade had too few observations for a reliable comparison, so the test she reports is the least-strong grade against the middle one — and all but two of the differences ran the wrong way, including all three of the statistically significant ones. Her conclusion states that the published strength estimates are not meaningful.
Consensus added nothing. Levels that more than one firm published had bounce frequencies close to those only one firm published, the differences were generally insignificant, and the single significant one favoured the unshared levels.
Then there is scope. Both of Osler's studies use FX quotes and FX order books, so neither establishes what happens at a level on the E-mini S&P or on crude. Her literature review reports two earlier tests of the other half of the convention — that price moves quickly once a level breaks — one finding it untrue on average for currencies outside strong trends, the other finding it held for daily moves of the S&P 500 stock index while the profits might not clear transaction costs. The same reservation applies as to candlestick patterns: the market a convention was tested in is part of the result.
Why would a price level matter at all?
One documented mechanism is the position of the orders already resting at the level. In a 2001 New York Fed staff report, Osler examined 9,667 conditional orders — stop-losses and take-profits — placed at one large currency dealing bank between 1 September 1999 and 11 April 2000, split 43% dollar-yen, 33% euro-dollar and 24% dollar-pound. About a third of them were executed.
Taking the last two digits of the requested rate on each order that executed, an even spread would put about 1% on every ending; the test she runs rejects that at far beyond the 0.01% level. The largest cluster sat at endings of 00, at about 8.7% of orders, with smaller ones at 50, at other endings in 0, and at 5.
Two asymmetries inside that clustering do the explanatory work:
- At the round number. On a rise into a level ending in 00, roughly 2.8% of all stop-loss buy orders trigger against roughly 10.5% of all take-profit sell orders, with similar average sizes on each side. The flow released at the level therefore tilts toward selling, and the rate has to fall far enough to draw buying out of the rest of the market. Osler calls this a natural if heuristic account of why rates stall there.
- Just beyond it. Stop-loss buy orders cluster above the round number rather than below it: 7.4% of them sit at endings from 90 to 99, against 14.4% at endings from 01 to 10. Stop-loss sells mirror that below the level, and take-profit orders show no such skew. Price that gets through the level runs into buying that was waiting on the far side.
The paper presents this as an explanation rather than a trading rule. Its value to a futures trader is conceptual: a level does something when orders are piled at it, so a level everyone can see is a level everyone's stop orders sit near.
What does this mean on a simulated funded account?
On an evaluation or a funded simulated account, a level is first of all a distance. The gap between your entry and the level you would trade against is a fixed number of ticks; multiply it by the tick value and the number of contracts and you have what that idea costs if it fails, before the question of whether price respects the level arises at all.
That number has to fit inside the account's loss line, which on our accounts sits a fixed distance below the highest end-of-day balance, never moves down, and liquidates the account the moment the balance touches it — the mechanics are in drawdown explained. The same idea at a wider tick distance is a different amount of money at risk, while the loss line has not moved an inch. Working that division through is the subject of our post on position sizing.
The order-clustering evidence adds one practical caution. In the book Osler studied, stop-loss buys bunched just above the round numbers rather than just below them, and stop-loss sells did the mirror image. That is one bank's book in three currency pairs, so it settles nothing about where a stop belongs on a futures contract. What it does show is that the obvious place can be a crowded one.
Frequently asked questions
Are support and resistance levels the same thing as pivot points?
No. A pivot point is arithmetic: Sierra Chart's default formula takes the prior period's high, low and close, averages the three for the pivot, and then derives each support and resistance line from the pivot and those same figures, so two traders using it over the same period draw the same lines. Support and resistance, in the manuals Osler reviewed, is a judgement about where buying or selling will concentrate, drawn from prior turning points, round numbers, retracement fractions or order flow. In her sample only about 30% of the levels two firms could have agreed on matched, at a five-point tolerance.
Do support and resistance levels work in futures markets?
Neither study described here answers that. Osler's bounce test measured published levels against intraday quotes for the mark, yen and pound from 1996 to 1998, and her order-book paper covers three currency pairs at one dealing bank in 1999 and 2000. Both are foreign exchange studies. Her review mentions an earlier test of the break-and-run claim on daily moves of the S&P 500 stock index, which is an index rather than a futures contract.
Why do round numbers keep showing up as levels?
Because that is where orders are. In the order book Osler studied, rates ending in 00 held about 8.7% of the orders that executed, against the roughly 1% an even spread would give, with secondary clusters at 50 and at other endings in 0. The analysts' own published levels show the same habit: more than 70% ended in 0 and 96% in 0 or 5.
If several analysts name the same level, is it stronger?
Not in the data. Osler compared levels published by more than one firm against levels published by only one and found the bounce frequencies close together, with the differences generally not statistically significant and the one significant difference favouring the levels nobody else had named. The firms' own strength ratings fared no better: the significant differences all ran opposite to the rating.
Sources
- CFTC — Futures Glossary (entries: Support, Resistance, Technical Analysis)
- Federal Reserve Bank of New York — Osler, 'Support for Resistance: Technical Analysis and Intraday Exchange Rates', Economic Policy Review, July 2000
- Federal Reserve Bank of New York — Osler, 'Currency Orders and Exchange-Rate Dynamics: Explaining the Success of Technical Analysis', Staff Report 125 (2001)
- Sierra Chart — Pivot Points Study (Formula Type, Formula Type 0)
- Sierra Chart — Technical Studies Reference: Swing High and Low
- Sierra Chart — Technical Studies Reference: High/Low for Time Period
- Quantower — ZigZag
- Lumen Futures — Help: Drawdown explained
Educational content about futures markets and simulated trading. Not investment advice, and not a solicitation to trade. Trading futures involves substantial risk of loss. Read the full risk disclosure.