RSI and MACD: calculation, interpretation and common errors

Two momentum tools, two different questions

RSI and MACD do not know the future. They summarize past price changes in different ways. RSI compares positive and negative movement on a bounded scale; MACD describes the separation of two averages and its relationship to a signal line. Decide what you want to measure before interpreting a color.

The examples here are deliberately hypothetical. The aim is to distinguish price continuing upward from momentum becoming less pronounced, and to separate both observations from a trading decision. These illustrations are not live signals, market data or evidence of strategy profitability.

How RSI is constructed

In the common 14-period version, each close-to-close change is separated into a positive gain and a loss expressed as a positive magnitude. Wilder smoothing produces average gains and losses. Their ratio is RS; RSI is 100 minus 100 divided by one plus RS.

A usual initialization takes the simple averages of the first 14 changes, requiring 15 closing prices. Subsequent averages multiply the previous value by 13, add the new observation and divide by 14. A rolling simple average is not the same smoothing method. Initialization and the amount of preceding history can also affect the early agreement between platforms.

If the smoothed average gain is 2 and average loss is 1, RS is 2 and RSI is approximately 66.67. For a subsequent gain of zero and loss of 3, the updated averages become 26/14 and 16/14. RSI is then approximately 61.90. This is one update from assumed starting averages, not a reconstruction of actual market prices.

If the smoothed loss is zero while the gain is positive, the mathematical RSI limit is 100; the reverse case tends to zero. If both are zero, the ratio is undefined. Check the implementation rather than inventing a directional-strength reading for a completely unchanged market.

The 70 and 30 levels are zones, not commands

Readings above 70 and below 30 are conventionally described as momentum extremes. RSI of 75 does not imply a 75% chance of a decline. RSI of 25 does not imply a 75% chance of a rise. The indicator is not a probability estimate.

In a persistent trend, RSI can remain elevated or depressed. Selling merely because it crossed 70 adds a reversal assumption not contained in the calculation. First decide whether the tool is a context filter, a pullback measure or a way to monitor weakening movement. Those are different jobs.

The 50 level relates to the balance of smoothed gains and losses; it does not prove that the price chart has no trend. Specify the timeframe and price source. A length of 7 or 21 is not inherently superior to 14: it changes the sensitivity and the question being asked.

MACD, its signal line and the histogram

With common 12, 26 and 9 settings, the MACD line subtracts the 26-period EMA from the 12-period EMA. The signal line is usually a nine-period EMA of MACD itself. The histogram subtracts the signal line from MACD. If the platform offers SMA and EMA alternatives, record the average type as well as each length.

Unlike RSI, MACD is not bounded between zero and 100. Its values are in price units. MACD of 2 in two differently priced assets does not establish equal momentum strength. Cross-asset comparisons need an explicit scale or normalization method.

A MACD/signal crossover is not the same event as MACD crossing zero. One compares MACD with its own average; the other reflects the ordering of the fast and slow averages. A positive histogram means MACD is above its signal, not that price must be rising at that instant.

A smaller histogram that is still positive

Suppose the fast EMA is 104 and the slow EMA is 102, giving MACD of 2. A signal of 1.5 gives a histogram of 0.5. In another hypothetical observation, MACD of 1.8 and signal of 1.6 give a histogram of 0.2. The gap has narrowed but remains positive. These are two arithmetic states, not a claimed pair of consecutive candles with a reconstructed EMA path.

Report the narrowing gap as a narrowing gap. Concluding that it is time to sell requires other rules that this subtraction does not supply. The time-series panels below use a synthetic price series and explicit calculations so that the line and histogram relationships can be inspected.

Divergence is a review prompt, not a reversal timer

A regular bearish divergence compares a higher price high with a lower indicator high at corresponding swings. A regular bullish divergence compares a lower price low with a higher indicator low. Define the swing-selection method first: unrelated comparison points can make almost any chart appear divergent.

A confirmed swing commonly requires later bars. In historical evaluation, timestamp the observation when it actually became identifiable, not at the earlier bar on which a marker was eventually drawn. Until then, it is a developing interpretation. Repeated divergences do not ensure that a trend will immediately end.

Agreement is not necessarily fresh evidence

Price above an average, RSI above 50 and positive MACD can all be related descriptions of the same preceding movement. Counting them as independent tests can create unjustified confidence. Give each tool a specific role and evaluate the final decision against structure, execution conditions and a tolerable loss boundary.

A practice template might use structure to describe direction, one momentum tool to measure movement quality, a reproducible observation condition, structural invalidation and position sizing. This is an example of process design, not a tested trading strategy or an instruction to transact.

Record the calculation contract

• Symbol, data venue and timeframe; exchange and session differences matter.
• Price source, RSI length and smoothing; MACD fast, slow and signal lengths and average types.
• Whether the candle is complete, and when any higher-timeframe value became available.
• Warm-up history and initialization method.
• Fees, slippage, invalidation criteria and conditions excluded from the process.

Matching the indicator's name is not enough to reconcile two outputs. Align these data and calculation choices before concluding that one platform is wrong.

A practical review exercise

Record ten completed-bar observations in a trending sample and a sideways sample. For each, separate what price did, what the indicator measured and what you additionally inferred. Then ask whether an actionable rule remains after removing the unsupported inference.

When evaluating a method, examine average gains and losses, drawdown, costs, sample size and untouched out-of-sample observations alongside win rate. Without that work, an attractive chart remains an educational illustration rather than evidence of an edge.