Definition and scope: what “Tema” refers to
Before collecting data, define what you mean by Tema in plain terms. In many moving-average discussions, “Tema” is used to describe a specific moving-average construction built from multiple moving-average stages. Your first data requirement is therefore the exact method description you will test: the formula, the number of stages, and the parameters (for example, the lookback length or other settings used inside the construction).
Next, separate stable mechanics from variable conditions. The mechanics are the deterministic part: the algorithm that takes input time series and produces an output series. Variable conditions include the market data used as input, the time granularity, any data adjustments, and the execution environment if you later relate outputs to trading-like outcomes. In this article we focus on assessment of the indicator/series construction itself, not on projecting future performance.
Data inputs: the time series and parameters you must specify
To assess Tema, you need at least these inputs:
-
The underlying price series (input time series). Decide which price field the method uses (commonly close, but the method definition must state it). You also need the instrument universe (for example, the specific FX pair) because the data values differ by instrument.
-
The timestamped sampling frequency. Specify the bar interval (for example, 1 minute, 1 hour, daily) and the timezone convention used in the dataset. Without this, you cannot reproduce results.
-
The parameter values. Provide the chosen lookback length and any other internal settings required by the Tema definition you adopted.
-
Data preprocessing rules. Document how missing values are handled, whether there are corporate-action adjustments (less common for FX, but data providers still differ), and how outliers are treated. These choices can change the output even when the algorithm is the same.
How does this work in practice? Tema is computed from its input series through the method’s calculation steps. If two sources disagree on timestamps, input price fields, or preprocessing, the resulting Tema series will differ even if the formula is unchanged.
Provenance and timeliness: where the data came from and when it was valid
Assessment also requires data provenance. You need to know:
- Source identity. Which dataset or provider supplied the input time series.
- Acquisition time vs. historical validity. Even for historical data, providers can revise backfills. Record the access date/version if available.
- Timeliness for any “current” claims. If you are assessing the latest computed values, timeliness becomes part of the dataset: when the data snapshot was taken matters.
A material quality check is to confirm that the dataset’s bar formation rules match your method’s expectations. For example, different providers can define bar open/high/low/close in slightly different ways or apply different session boundaries. Those differences can shift the input series and therefore the Tema output.
Evidence and quality checks: how to verify the assessment is consistent
Use a verification approach that checks both the mechanics and the data quality:
- Reproducibility test. Recompute Tema using the same definition, parameters, and input dataset. If you cannot reproduce the output within a small tolerance, the assessment has a data or implementation problem.
- Cross-provider comparison. If you use two independent data feeds, compare whether the discrepancies are consistent with known differences (timestamps, price field, missing data handling). Large unexplained differences are a rode vlag.
- Boundary-condition check. Confirm how the method behaves near the start of the series, where insufficient prior data can force different initializations.
- Calculation trace. When possible, capture intermediate series (the method’s internal moving-average stages) so you can locate where differences enter.
Klaarcriterium: you can state, with clear documentation, (1) the exact Tema definition, (2) the precise input series and parameter settings, (3) the dataset provenance and timestamping, and (4) the quality checks you applied.
Limitations and risks: what can go wrong and what you should not assume
One material limitation is sensitivity: moving-average outputs depend strongly on the chosen input data, the sampling interval, and the method’s initialization. Therefore, historical relationships do not establish future results.
Another limitation is non-stationarity. Market regimes can change, and an algorithm that describes one period may behave differently in another period, even when computed correctly.