Market Cycles and Seasonality for CMT Candidates: Measure Before You Forecast

Scope of the current curriculum

The 2026 CMT Program Guide includes cycle foundations at Level I: period, amplitude, phase, harmonics, and other cycle concepts. Level II objectives cover applied and seasonal cycles, detrending, dominant-cycle tools, left and right translation, inversion, and related methods. It also includes Elliott Wave. The public guide mentions notable economic cycles, but it does not name Kitchin, Juglar, or Kondratieff in the objective text. Those historical labels can provide context, yet should not be presented as individually verified 2026 exam requirements without checking the assigned official chapter.

A cycle hypothesis says a measure tends to oscillate around a reference over time. A seasonal hypothesis says a pattern aligns with a calendar unit such as month or quarter. Markets can exhibit both, neither, or a relationship that changes over time. Apparent repetition on one chart can arise by chance or by selecting a convenient window after the fact.

Measure the components

A simple periodic model can be described in plain text as price component = baseline + amplitude × sine(2π × time/period + phase). The period is the time for one full oscillation. Amplitude is its distance from baseline to peak. Phase says where the wave sits in the cycle at a given time. If the baseline is 100 and amplitude is 8, the idealized oscillation runs between 92 and 108. A 20-week period would place ideal peaks 20 weeks apart if the model remained stable.

Real prices also trend and jump. If a stock rises 30% across a year, a seasonal average of raw prices may confuse the trend with the calendar effect. Detrending or working with returns can help, but the transformation must be specified. A centered moving average may help estimate a local baseline in a historical study, yet it uses future observations around a center point; using that smoothed value in a real-time backtest can introduce look-ahead bias.

A worked seasonal calculation

Suppose an asset's returns for a particular month across five years are +2%, +3%, −1%, 0%, and +1%. The arithmetic monthly average is (2 + 3 − 1 + 0 + 1)/5 = +1%. Four of five observations are nonnegative, but the sample is tiny. The median is +1%. If transaction costs would total 0.4% for a round trip, the naive mean net result is about 0.6% before tax and other slippage, assuming the same trade mechanics in each year. This is a descriptive calculation, not evidence of a robust seasonal edge.

Compare this month with all other months under the same years and market regime. If you inspect twelve months and report only the best one, selection bias grows. A calendar effect may also reflect one unusual year. Record the range of outcomes and whether the effect persists in a later period. If the latest three years are −1%, 0%, and +1%, the recent average is 0%, which should temper confidence in the five-year +1% estimate.

Translation and inversion

Cycle translation describes the location of a peak within a cycle. In an ideal 20-week cycle beginning at a low, a midpoint peak occurs near week 10. A peak at week 15 is right-translated, often interpreted as strength because gains persist later into the cycle. A peak at week 6 is left-translated, often read as weakness. These are contextual interpretations; the cycle boundary itself must be identified without using future bars if the analysis claims to be real-time.

An inversion occurs when a projected high appears as a low or vice versa. It reveals that a fixed phase projection can fail. Do not relabel every miss as an inversion and still claim the model was correct. A cycle model needs a predeclared tolerance for timing and sign; outside that tolerance, count a forecast failure.

Harmonics are cycles with related periods. A 10-week oscillation can be a harmonic of a 20-week one. Their combined waveform can have uneven peaks and troughs. A complex-looking chart may therefore invite several plausible period choices. A spectrum or periodogram can help identify candidate frequencies, but unstable markets and short samples can produce spurious peaks. Repeat the analysis on later data and alternative intervals.

From cycle idea to a decision

One illustrative rule is to consider a seasonal tilt only if the effect is positive in a predeclared historical window, remains positive in a holdout window, and the asset's price is above its own long-term trend filter. If any condition fails, do not apply the tilt. An invalidation rule could reduce exposure if price closes below a marked support level, rather than waiting for a predicted cycle date. This is a research framework, not a claim that seasonality generates returns.

Calendar boundaries can vary by market. Holiday schedules, fiscal cycles, and contract rolls influence data. A commodity's seasonal supply pattern may have a causal narrative, but the tradable return pattern can disappear after storage costs, hedging, and anticipation. Equity calendar labels without a plausible mechanism deserve even more skepticism.

Common mistakes and self-check

Do not infer a reliable period from two peaks. Do not use a centered smoothing value in a backtest as if it had been available at the center date. Do not confuse an average month with every year. Do not present named historical economic cycles as a public 2026 objective without checking the curriculum. A cycle projection should name time units, tolerance, and invalidation.

Self-check: Peaks occur at weeks 4, 16, and 28. What is the observed peak-to-peak interval, and can you forecast week 40 with certainty? Answer: Both gaps are 12 weeks, so the sample suggests a 12-week candidate period. Three peaks provide little evidence of stability; week 40 is a hypothesis to test, not a certain turning point.

Next, compute one calendar month's return for at least five years, include all observations, and write how the result changes when the strongest year is removed.