This page explains how the current MACD research branches were selected for live paper observation.
The active branches were not chosen by manual chart selection or by looking for a visually attractive historical result.
They were selected through a structured historical screening process using the same general MACD Sample-style logic across multiple currency pairs, timeframes, and maximum holding periods.
The purpose of this process was not to prove future profitability.
The purpose was to identify configurations that showed enough historical potential to justify the next research stage.
Research Objective
The main research question was:
Can a transparent momentum-based MACD model produce sufficiently stable positive paper results across different FX pairs?
The project was not looking for one isolated historical winner.
It was looking for several candidates with different currency structures so that the same general logic could be observed across different markets.
Starting Point
The screening started with a transparent MACD Sample-style trading model.
The logic was converted into a Python historical test so that the same rules could be applied consistently across all tested configurations.
The model used predefined:
- entry conditions;
- exit conditions;
- stop-loss and take-profit logic;
- maximum holding time;
- trade tracking;
- performance measurement in R.
The same general logic was then tested across different:
- currency pairs;
- timeframes;
- maximum holding periods;
- historical market periods.
Metrics Used in Historical Screening
The screening process considered several performance and robustness metrics.
- Number of trades — whether the sample was large enough to provide useful information;
- Win rate — the percentage of profitable trades;
- Gross positive R — the total R generated by profitable trades;
- Gross negative R — the total R lost in unprofitable trades;
- Net R — the final result after combining positive and negative R;
- Average R per trade — the average result generated by each trade;
- Positive week rate — the proportion of profitable weeks;
- Worst week — the weakest weekly result in the test;
- Maximum drawdown — the largest cumulative decline during the test;
- Result after removing the strongest week — a robustness check designed to reduce dependence on one exceptional period.
No single metric was treated as sufficient by itself.
A high Net R could still be weak if it depended on one unusually strong week, contained excessive drawdown, or was produced by too few trades.
Why the Strongest Week Was Removed
A trading model can appear stable when most of its historical profit came from one exceptional market period.
This creates concentration risk.
For that reason, the screening included a test in which the strongest week was removed from the result.
The purpose was to answer a simple question:
Does the candidate still look acceptable without its best week?
A version that remains positive or reasonably stable after this adjustment may have a more balanced historical profile.
A version whose result collapses after removing one week may be too dependent on an isolated event.
This test does not eliminate overfitting and does not predict future results. It is only one additional robustness check.
Why Several Currency Pairs Were Selected
The project did not want to rely on one currency pair only.
A logic that performs well on one pair may be reacting to a specific market structure rather than demonstrating broader potential.
For that reason, the final research basket includes:
- a EURJPY branch;
- an AUDCAD branch;
- a GBPUSD branch.
These pairs represent different currency combinations and different market behavior.
The goal is to determine whether the MACD logic has wider potential or whether positive historical behavior was limited to one specific pair.
EURJPY MACD V Base
Selected configuration:
- Symbol: EURJPY;
- Timeframe: M15;
- Maximum holding period: 24 hours;
- Status: Base control version;
- Mode: PAPER_SIM_ONLY;
- Orders sent: NO.
EURJPY MACD V Base was the first strong MACD branch identified during historical screening.
The historical test showed meaningful positive potential for the MACD Sample-style logic on EURJPY using the M15 timeframe.
This version was selected as the control model for the EURJPY MACD branch.
The word Base means that the version keeps the original selected structure for that branch.
It is not assumed to be the final robot.
Its purpose is to provide an unchanged reference version against which later modifications can be compared.
Why EURJPY MACD V.30 Was Added
EURJPY MACD V.30 was added because the historical audit showed both positive potential and a clear weakness related to maximum holding time.
The Base version uses a 24-hour maximum holding period.
The modified V.30 version uses a 16-hour maximum holding period.
This was not a random change.
Only one major variable was changed:
MAX_HOLD was reduced from 24 hours to 16 hours.
The purpose was to test whether a shorter holding period could:
- reduce weak time-based exits;
- improve Net R;
- reduce drawdown;
- produce a smoother cumulative R path;
- close weak trades earlier without removing too many profitable trades.
Both versions were launched in parallel so that they could be observed under the same live paper market conditions.
EURJPY MACD V Base remains the unchanged control version.
EURJPY MACD V.30 is the controlled modification.
The purpose is not to assume that V.30 is better.
The purpose is to measure whether the shorter holding period actually improves the live paper result.
AUDCAD MACD V Base
Selected configuration:
- Symbol: AUDCAD;
- Timeframe: M30;
- Maximum holding period: 36 hours;
- Status: Base control version;
- Mode: PAPER_SIM_ONLY;
- Orders sent: NO.
AUDCAD MACD V Base was selected through a separate historical screening of non-major currency crosses.
The goal was to identify a candidate that did not use EUR, JPY, GBP, or USD.
AUDCAD on the M30 timeframe with a 36-hour maximum holding period produced the strongest historical profile within that non-major screening.
This made AUDCAD useful for diversification inside the research basket.
It provides exposure to AUD and CAD behavior rather than repeating the same currency structure already represented by EURJPY or GBPUSD.
AUDCAD MACD V Base remains an unchanged reference version until there is a clear evidence-based reason to create a modification.
GBPUSD MACD V Base
Selected configuration:
- Symbol: GBPUSD;
- Timeframe: M30;
- Maximum holding period: 12 hours;
- Status: Base control version;
- Mode: PAPER_SIM_ONLY;
- Orders sent: NO.
GBPUSD MACD V Base was selected as a major-pair candidate outside the EURJPY branch.
Historical testing showed that GBPUSD on the M30 timeframe with a 12-hour maximum holding period had a strong enough profile to justify live paper observation.
GBPUSD adds a different market structure from both EURJPY and AUDCAD.
This helps test whether the MACD logic has broader potential across:
- a EUR/JPY cross;
- a non-major AUD/CAD cross;
- a major GBP/USD pair.
GBPUSD MACD V Base remains an unchanged reference version until further evidence justifies a controlled modification.
Why the Base Versions Must Remain Unchanged
A Base version serves as the control model for its research branch.
If the Base version is repeatedly changed, it becomes impossible to determine whether a later result was caused by:
- market conditions;
- a new filter;
- a changed holding period;
- a different entry rule;
- a different exit rule;
- several changes introduced at the same time.
For this reason, the preferred method is:
keep the Base version unchanged, create a separate modification, change one important variable, and run both versions in parallel.
This creates a clearer comparison and reduces the risk of drawing conclusions from mixed changes.
Why Historical Testing Is Not Enough
Historical screening is only the first selection stage.
A historically positive result may fail during live paper observation.
Possible reasons include:
- different spread behavior;
- signal timing differences;
- data quality differences;
- overfitting;
- unstable trade distribution;
- dependence on one market regime;
- weak maximum holding exits;
- performance concentration in a small number of trades.
For that reason, no Base version is treated as a finished product after historical screening.
Each selected version must first be observed in live paper mode.
Live Paper Observation Questions
During live paper observation, the project examines questions such as:
- Are signals generated as expected?
- Is the trade frequency similar to the historical model?
- Does Net R remain positive or reasonably stable?
- Is drawdown controlled?
- Are losses concentrated around MAX_HOLD exits?
- Does the cumulative R path remain stable?
- Does one version depend on a small number of unusually strong trades?
- Should the version continue unchanged?
- Should one controlled modification be created?
- Should the branch be paused or rejected?
Current Safety Status
All current MACD branches operate in research-only paper mode.
PAPER_SIM_ONLY
orders_sent=NO
No live orders are sent.
The current versions are not downloadable Demo Robots.
Paper trading does not reproduce all real trading conditions, including slippage, rejected orders, execution delays, spread spikes, and broker-specific behavior.
For this reason, a positive paper result would still require additional review before any demo-stage decision.
Current Research Basket
The active MACD research basket currently contains:
- EURJPY MACD V Base;
- EURJPY MACD V.30;
- AUDCAD MACD V Base;
- GBPUSD MACD V Base.
The three Base versions test the same general MACD research direction across different markets.
EURJPY MACD V.30 tests one controlled change against its Base control version.
Research Decision Principle
A historically selected candidate does not automatically become a Demo Candidate.
A Base version may:
- continue unchanged;
- receive a controlled modification;
- remain in observation;
- be paused;
- be rejected;
- move toward demo review after stronger evidence.
The decision must be based on measured behavior rather than on a single profitable week or isolated result.
Related Research Pages
Important Notice
This page documents the research selection process only.
Historical screening, live paper results, demo results, weekly reports, and past performance do not guarantee future results.
FX Trading Robot Lab does not provide investment advice, financial advice, managed account services, copy trading, trading signals, or guaranteed trading results.