EMA vs SMA: Check the Calculation, Weights and Starting Value

You have a moving-average number on one screen and a different number in a spreadsheet. Before deciding that either formula is wrong, identify the exact series and the starting rule. A value labelled EMA 3 does not describe all the choices needed to reproduce it. This guide ends with a numerical audit: the input prices, each available average, the weights behind a selected row and a record of anything still unknown.

Use one instrument, one price definition, one unit and completed periods of equal length, ordered oldest to newest. Close can mean a bid close, ask close, last-trade close or another supplied reference. Those are different inputs. Record the provider, time zone, session boundaries and whether the series is adjusted. For an instrument with corporate-action adjustments, an adjusted historical close can differ from an unadjusted one.

The checker accepts a sequence of values, not timestamps or a market feed. It cannot verify that your rows really are closed, equally spaced periods. Establish that from the source before pasting. Repeated equal prices are legitimate observations; a missing period is a different issue and must not be represented by an invented zero.

SMA stands for simple moving average; EMA stands for exponential moving average. For row t, the simple moving average is SMAₜ = (pₜ₋ₙ₊₁ + … + pₜ)/N. It becomes available at row N, after N observations exist. Rows 1 through N − 1 have no value under this full-window convention. The MetaQuotes moving-average reference describes the arithmetic mean and the exponential recurrence; its separate smoothed-moving-average section should not be mistaken for an EMA initialization rule.

For this guide, α = 2/(N + 1) and β = 1 − α. After initialization, EMAₜ = αpₜ + βEMAₜ₋₁. The CME explanation gives this multiplier and allows an initial SMA or a previous close as a starting value. The TradingView EMA explanation also describes exponential updating with a smoothing factor of 2. Those method descriptions do not establish one universal seed for every platform, history length or setting.

The first-price option here means the first price in the supplied series. It does not pretend that you have supplied the platform’s entire earlier history or its hidden previous EMA. Both options use the same α. Choosing a seed is a calculation convention for reconciliation, not a recommendation about which indicator to trade.

When N = 1, α = 1 and β = 0. SMA and both EMA variants equal the current price at every row. With N equal to the series length, SMA and the first-SMA EMA appear only on the final row. If the series has more than one price, the first-price EMA is available earlier because its initialization needs only one value.

Take the invented series 100, 110, 90, 110, 100, expressed as USD prices for an unnamed teaching instrument. Each row represents a completed equal-length period in the exercise; these are not market observations. Set N = 3, so α = β = 1/2. Keep fractions or full numerical precision until the final display.

The first full SMA is (100 + 110 + 90)/3 = 100. At row 4, the window is 110, 90, 110, giving 310/3; at row 5, it is 90, 110, 100, giving 100. The oldest price leaves the SMA window completely when a new price enters.

For the first-SMA EMA, row 3 starts at 100. Row 4 is 0.5 × 110 + 0.5 × 100 = 105; row 5 is 0.5 × 100 + 0.5 × 105 = 102.5. For the first-price EMA, row 1 is 100; successive values are 105, 97.5, 103.75 and 101.875. The two methods have used the same prices and multiplier but different initialization.

The last first-SMA EMA minus the last first-price EMA is 102.5 − 101.875 = 0.625 USD. It is a difference between two average prices, not 0.625%, a return or a trading gain. At row 3 the seed difference is 2.5; two further updates multiply it by β twice, so 2.5 × 0.5² = 0.625. This isolates the effect of initialization without changing the data.

A useful reconciliation shows why the output has that value. For SMA, the last N prices each have weight 1/N and all older prices have weight zero. For EMA, a finite supplied history has explicit weights that include the starting value. Omitting that initial contribution would leave the weights short of one and understate the average.

For a first-SMA seed at row t ≥ N, each of the first N prices has weight β^(t − N)/N. Each later price pⱼ, with j > N, has weight αβ^(t − j). For a first-price seed, p₁ has weight β^(t − 1), while each pⱼ with j ≥ 2 has weight αβ^(t − j). These are weights for the chosen row, not for observations that have not occurred yet.

Multiply each price by its weight and add the contributions. The first-SMA EMA at row 5 is 100/12 + 110/12 + 90/12 + 110/4 + 100/2 = 102.5 USD. The first-price EMA is 100/16 + 110/16 + 90/8 + 110/4 + 100/2 = 101.875 USD. Weights are dimensionless; contributions and the final average use USD.

The inspector shows every observation up to the selected row, including any zero SMA weight, and the contribution totals. When the selected method is not available yet, its weights and average remain unavailable. For N = 1, the current price has weight one and all earlier prices have weight zero; there is no need to interpret an ambiguous zero-to-zero power.

Paste one price per line, oldest first, without a header. A point or comma may separate decimals, but commas must not separate observations: 100,5 is one value. Do not paste a CSV table, currency symbols, thousands separators or exponent notation. The input CSV download contains a header and the reference values; copy only its price cells into the form. This textarea is not a CSV importer.

The supported range is 1–240 positive prices, each no greater than 1,000,000,000 and with at most 12 decimal places. Each line is limited to 80 characters and the complete input to 20,000 characters. One final line break is allowed. Leading or internal blank lines are errors because a missing observation must not be silently discarded. Zero, negative values, non-finite tokens and malformed entries are rejected.

Choose integer N from 1 to the number of prices, a price-unit label of 1–24 characters, and one of the two explicit EMA seeds. The unit label starts with a letter or number, followed by letters, numbers, spaces or . _ - /; it labels the data and does not convert currencies. Calculate to see both EMA series and the SMA, with the selected EMA used in the chart and weight breakdown.

Changing prices, N, units or the seed hides the previous calculated result and its CSV until you calculate again. Selecting another inspection row only navigates the already calculated result; it does not erase it. The reference, flat and spike presets are synthetic demonstrations, not recommended settings or tested strategies.

Exact fractions are retained throughout. Repeating decimals and display-shortened values are clearly marked as approximations; the display uses 18 significant digits and does not feed rounded values back into the recurrence. Inspect the exact export when more digits matter. The full reference table and static calculation CSV remain available without JavaScript, together with the input CSV and verification protocol.

The component keeps entries in page memory and does not send them to a server or write persistent storage. Site-wide analytics are separate. Without JavaScript the interactive controls are disabled, while the complete fixed example and downloads remain readable.

Put the five reference prices in B2:B6, oldest first. Set H2 to 3 for N and H3 to =2/(H2+1) for α. Use column C for SMA, D for the first-SMA EMA and E for the first-price EMA. Keep the first two cells in columns C and D empty; entering zeros would invent unavailable averages.

For a different N, move the first full SMA and first-SMA seed to the row containing the Nth price, then extend the same logic. Do not merely change H2 while leaving a three-price window in the formulas. The window length, first available row and seed range must all agree with N.

SUM is written in English here; your spreadsheet may use a localized function name. Decimal separators also depend on locale. Formula references and arithmetic mean the same thing. Keep formatting separate from calculation: do not insert ROUND into every update. Ordinary spreadsheets have finite numerical precision, so their displayed last digits need not equal an arbitrary-length exact fraction.

The downloads separate exact numerators and denominators from decimal displays. Import exact fraction and long-integer fields as text to avoid automatic date conversion or lost digits. A useful check is agreement of the reference values, weight sums and weighted totals; a screenshot of a rounded chart alone cannot establish the full calculation.

Compare from the earliest row where both calculations have a value. If the input prices match but the EMA already differs at the seed, record the initialization difference first. With the same α and the same subsequent prices, an existing difference is multiplied by β at every update. That explains gradual convergence; it does not justify claiming that an unknown starting value has been recovered.

• Different history: the first price you pasted may be much later than the first price used by the other system. Record what earlier data or previous EMA is missing.
• Different close: verify bid, ask, last-trade or other source definitions instead of assuming that all fields called close are equal.
• Different periods: check time zone, trading session, resampling boundaries and whether all rows are completed periods.
• Adjusted data: document adjustment conventions where they apply; do not mix adjusted and unadjusted histories.
• Different calculation policy: confirm N, multiplier, initialization and intermediate rounding. A label such as EMA 3 does not settle those choices.

The backtesting guide addresses the separate problem of using information only after it becomes available. Here the narrower check is that a later price cannot change an already calculated earlier average. The calculator does not simulate orders, estimate strategy performance or verify a terminal’s historical feed.

Use the downloadable protocol to retain the source, price definition, unit, interval, session, observed period range, seed and unresolved questions. If the history or convention is unavailable, report the gap rather than adjusting numbers until two rounded outputs happen to agree.

At row 5 of the N = 3 example, the newest price has weight 1/2 in both EMA variants, while its SMA weight is 1/3. This explains one concrete difference in weighting. EMA also keeps declining contributions from earlier history; SMA discards observations outside its current window. Those definitions alone do not identify a universally better period or a profitable crossover.

Because the weights here are non-negative and sum to one, the average lies within the range of the prices that contribute to it. A constant positive series gives that same constant once each method is available. Multiplying all prices by a positive factor multiplies every available average by the same factor; adding a constant shifts them by that constant, provided the new inputs remain in range.

A moving average smooths price levels. It is not a measure of the dispersion of price changes, which is a different task covered by the Forex volatility guide. Neither this chart nor its synthetic spike preset measures trading risk, predicts a future close or recommends an entry or exit.

The reproducible outcome is a calculation record: exact inputs and conventions, available SMA and EMA values, the selected row’s weights and a list of missing evidence. Exact arithmetic can establish what follows from those inputs. It cannot establish that the inputs match a particular platform or that the indicator has predictive value.

Does EMA use only the last N prices?

No, except for N = 1. Here N sets α = 2/(N + 1). When N > 1, earlier observations continue to contribute after initialization with declining weights. SMA uses only its current N-price window. For N = 1, both methods use only the current price.

Why can one EMA start earlier than the SMA?

The first-price seed needs only p₁, so it is available at row 1. The first-SMA seed needs N prices and starts at row N. Earlier unavailable values are not zeros.

Which EMA starting rule is correct?

Both are explicitly defined calculations. For reconciliation, use the rule and history of the series you are checking when they are known. This guide does not claim a universal platform seed or choose a better trading indicator.

Can I remove blank rows before pasting?

Only after identifying what they mean in the source. A blank caused by a missing period is not an ordinary formatting gap. The checker rejects leading and internal blanks rather than silently compressing time or inserting zeros; one final line break is allowed.

Should I round every EMA update in Excel?

Not for comparison with this exact model. Keep intermediate calculations at the spreadsheet’s available precision and format the final display. If another system deliberately rounds each step, document that as a separate calculation convention.

Will the result exactly match my trading platform?

Only identical inputs and conventions can support that expectation. Check history, close definition, session boundaries, N, seed and rounding first. The calculator does not fetch platform data or certify a platform match.