A moving average that reacts less to noise also reacts less to new information. That trade-off sits underneath the appeal of a slow trend line. It deserves a measurement, because “fewer fakeouts” can mean two very different things: avoiding expensive reversals, or noticing a genuine change later.
APEX’s founder uses the 233 smoothed moving average alongside the 21 and 55 exponential moving averages. The reason is personal and explicit: 233 is a Fibonacci number, and the founder has found the line steadier than a 200-period alternative in practice. That is a useful hypothesis to investigate. It does not establish that Fibonacci settings outperform, or that everyone should adopt the same chart.
This study quantifies the filter itself. In a controlled experiment, the 233 SMMA needs 162 completed bars to absorb half of a permanent change. A 200 EMA needs 70. A 200 SMMA needs 139. Those differences are substantial, but their trading value depends on what happens while the slower line waits.
First, define what “the 200” means
The period number is only part of the indicator. A 200 simple moving average, 200 EMA and 200 SMMA distribute information differently. The founder’s original reference to “the 200” did not identify the averaging method, so this study keeps all three alternatives visible.
MetaQuotes documents an SMMA that starts from a simple average and then updates recursively. TradingView’s documentation specifies the conventional EMA coefficient as 2/(N+1), and its RMA smoothing coefficient as 1/N. These formulas make it possible to compare the response before introducing any market, entry rule or performance claim. [1, 2]
For the recursive filters, each new value equals alpha × current input + (1 − alpha) × previous average. The 233 SMMA assigns about 0.429% to the new input. The 200 EMA assigns about 0.995%. Changing from one to the other alters both the length and the smoothing method. It is a larger change than the labels “200” and “233” suggest.
The 200 SMA gives each of its latest 200 observations equal weight. It removes the oldest observation when the next one arrives. A recursive average instead retains a diminishing influence from older history. The experiment below makes those different memories visible.
The permanent-change test
Imagine an input that has always been zero. At bar one it jumps to one and stays there. Every average begins at zero. This deliberately simple experiment isolates responsiveness: no trend selection, favorable entry date or price-data vendor can determine the result.
For a recursive average, the fraction of the change absorbed after k observations is 1 − (1 − alpha)^k. For a 200 SMA it is min(k/200, 1). APEX calculated 800 bars and checked the direct formulas against a separate bar-by-bar recurrence. The results below are the first whole completed bars reaching each threshold.
| Filter | 50% absorbed | 90% absorbed | Response after 12 bars |
|---|---|---|---|
| 200 EMA | 70 bars | 231 bars | 11.31% |
| 200 SMA | 100 bars | 180 bars | 6.00% |
| 200 SMMA | 139 bars | 460 bars | 5.84% |
| 233 SMMA | 162 bars | 536 bars | 5.03% |
| 465 EMA, matched initialization | 162 bars | 536 bars | 5.03% |
“162 bars” is a filter response time. It is not a prediction that an actual crossover will occur 162 bars after a reversal. Real price paths, the other line in a crossover and the initial state all affect that event.
On an uninterrupted 24/7 market, 162 four-hour bars span 27 days; 162 eight-hour bars span 54 days; 162 daily bars span 162 days. Session-based markets have different calendar spacing. A slow line on a daily chart carries a very different decision horizon from the same setting on an intraday chart.
The short-shock test
Now let the change last only 12 bars before the input returns to zero. The 200 EMA moves 11.31% of the way toward the temporary level by bar 12. The 233 SMMA moves 5.03%. That explains a real mechanical benefit of stronger smoothing: a brief disturbance moves the line less.
It also exposes a cost. Once the input returns to zero, the recursive average retains a fading memory of the disturbance. The 200 SMA responds differently: its 12 elevated observations remain inside the window until they begin to roll out, creating a flat stretch before the response falls away. A chart can look stable because of its weighting rules, even while the underlying situation has already changed.
A smaller movement in the line is not a measured reduction in losing trades. Price can repeatedly cross a nearly flat average. A slower filter can also leave a position exposed through a sustained decline. The relevant outcome is what the complete decision rule does with that information.
The control that separates the number from the formula
A conventional 465 EMA has coefficient 2/(465+1) = 1/233. With identical input and identical initial value, it follows exactly the same recurrence as the 233 SMMA. APEX’s step and pulse outputs coincide at every calculated bar.
This control is useful because it removes the label from the argument. A trader can reasonably prefer the 233 setting. The mathematics does not require a special Fibonacci effect to produce its response.
Platform plots can still differ if their first values, available history, source prices or missing-bar rules differ. NIST discusses how initialization influences exponential smoothing, particularly with small coefficients. [3] For these experiments, the constant prehistory makes the initial state identical by construction. For a market comparison, initialization must be checked rather than assumed.
There is another precise distinction. In the idealized case of independent, equal-variance input noise around a fixed level, a recursive filter passes a variance fraction alpha/(2 − alpha); an N-period SMA passes 1/N. Those are sums of squared weights, calculated here by APEX. The 233 SMMA’s ratio is 1/465, compared with 1/200 for both the 200 EMA and 200 SMA. This is a mathematical noise model, not an assumption that real prices are independent noise or a claim about portfolio risk.
How this fits the 21 / 55 / 233 framework
The shorter pair and the slow reference answer different questions. A completed 21/55 EMA cross records a change in their relative ordering. Price’s position against the 233 SMMA provides slower context. Combining them can organize a review; it does not turn correlated information from the same price series into independent confirmations.
For research, APEX proposes the following conditional interpretation. These states are descriptive hypotheses, with no official model weight or automatic trading instruction attached.
- Momentum weakens while the slow context holds: the 21 EMA crosses below the 55 EMA, but the completed close remains above the 233 SMMA and that SMMA is higher than five bars earlier. Treat this as a mixed state. A later 21/55 bullish cross while the close remains above the slow line ends the momentum-warning episode.
- Weakness broadens: 21 EMA is below 55 EMA, the close is below 233 SMMA, and the SMMA is lower than five bars earlier. This satisfies the proposed broader-weakness definition. A completed close back above the slow line breaks that exact state; full constructive alignment requires the opposite ordering and a rising slow line.
- Repeated reversals dominate: the price/slow-line relationship flips back and forth within the defined evaluation horizon. Log every occurrence and its cost. Do not relabel failures after seeing the next move.
The five-bar slope comparison is an explicit research choice, not a discovered optimum. Evaluate four-hour, eight-hour and daily bars separately. TradingView documents that unfinished higher-timeframe values can change until their bars close. [4] A daily condition must therefore use a completed daily observation available at the decision time.
A fair test of “fewer fakeouts”
The next market study needs a definition that can lose. Here is the proposed protocol, frozen in this review package before any price-results are produced.
- Identify the inputs. Record symbol, venue, quote currency, session calendar, source field, timezone and publication rights. Use one identical, gap-audited dataset for every comparator. Keep gold futures rolls and crypto spot observations in separate experiments.
- Hold everything else fixed. Compare 200 SMA, 200 EMA, 200 SMMA and 233 SMMA, with 465 EMA as a matched-coefficient implementation control. Use the same 21/55 pair, sizing, exposure limits and execution assumptions. Do not optimize each alternative until its equity curve wins.
- Define an event. For the primary filter diagnostic, a completed close changes from at-or-below a line to above it, or from at-or-above to below it. A reverse cross within ten completed bars is a whipsaw episode. Allow one active episode per comparator; close it at the first reverse cross or after ten bars. Count a reversal as the end of that episode, not the simultaneous start of another. Report censored episodes separately.
- Measure frequency and consequence. Report episodes, reversals, exposure time, turnover and adverse movement. For a separately specified executable long/cash test, enter at the next bar’s open after an upward cross and exit at the next open after a downward cross. Deduct documented fees and slippage; report net return, drawdown and time invested alongside an equivalent hold benchmark. This long/cash diagnostic is not a replica of the founder’s discretionary trading.
- Protect the unseen period. Specify chronological development and evaluation dates before running results. Remove evaluation events that use future bars beyond the cutoff, and prevent event horizons crossing a split from leaking outcomes into selection. Record every attempted variant, including unsuccessful ones.
The choice of ten bars is a starting definition, not a universal truth. Five- and twenty-bar windows can be reported as predefined sensitivity checks, without selecting the best one afterward. Because market observations cluster, confidence estimates should preserve time dependence; a large trade count alone is not a large number of independent experiments.
Bailey, Borwein, López de Prado and Zhu show why repeated backtest selection deserves explicit scrutiny. [5] Their work does not establish anything about this 233 setting. It strengthens the case for recording the search process and resisting a winner chosen from many conveniently forgotten trials.
What would change the conclusion?
The mechanical result is narrow and reproducible: under identical initialization, stronger smoothing reduces the short-pulse response and slows adjustment to a lasting shift. Different formulas or inputs would require a different calculation.
The trading hypothesis remains open. Evidence would support the 233 preference if unseen, cost-adjusted results show a useful reduction in damaging reversals without an unacceptable increase in delayed exits or missed participation. Evidence against it would be fewer crossings but worse net outcomes, deeper drawdowns, or an apparent advantage that disappears outside the selected market period. If the 465 EMA control differs materially after inputs and initialization are matched, investigate the implementation before interpreting performance.
A smoother line can be a useful part of a trader’s process. The test is whether its delay fits that process. Education, position sizing and execution still determine what happens after a line moves.
Sources and research notes
Primary documentation and paper checked October 9, 2026. Calculations, experiment design and chart interpretation are original APEX work. This is a new response-model study following the October 8 moving-average article; that earlier article remains unchanged.
- MetaQuotes: Moving Average — SMA and SMMA definitions and initialization.
- TradingView: PineJS Utility Functions — EMA and RMA coefficients.
- NIST/SEMATECH: Single Exponential Smoothing — exponential weighting and initialization; this study uses a contemporaneous input update.
- TradingView: Repainting — completed and developing higher-timeframe values.
- Bailey, Borwein, López de Prado and Zhu: The Probability of Backtest Overfitting, February 2015 revision — strategy-selection risk.
Data limit: no usable current OHLC dataset was obtained for this study. A Coinbase BTC-USD candle request returned HTTP 403 on October 9. That access result is not evidence of a market condition. The market backtest requires a verified dataset with applicable use rights; the completed mathematical experiment does not.
Credits: research, calculations and charts: APEX Intelligence. The 233 preference is attributed to the APEX founder. Original cover illustration created with AI for APEX; decorative lines are not price data. No private trade records, proprietary engine rules or official scenario weights are disclosed.
Not financial advice. Always trade with a plan and proper risk management.