Trading Research

Phase 4 — Turn Indicator Observations into Testable Trading Research

Technical indicators can help us describe market behavior, but an observation is not yet evidence of a trading edge.

Phase 4 begins where Phase 3 ends. Instead of asking only how an indicator is calculated, we now ask how to turn an observation into a precise, reproducible experiment. 

indicator observation
→ define the state
→ define the event
→ freeze the rule
→ separate signal time from execution time
→ measure future outcomes
→ compare with a baseline
→ test uncertainty
→ build a backtest
→ validate the result

How Phase 4 Is Organized

A backtest should not begin with one large strategy script. It should begin with small research questions that can be checked one at a time. 

Rule Definition
→ What exactly is the event?

Forward Outcomes
→ What happened after the event?

Baseline Comparison
→ Was the outcome different from ordinary market behavior?

Backtest Mechanics
→ How would the rule have been executed through time?

Robustness & Validation
→ Does the result survive costs, parameter changes,
  different periods, and unseen data?

The goal is not to find a profitable-looking chart as quickly as possible. The goal is to build an experiment whose logic, timing, assumptions, and results can be reproduced.

1. From Indicator to Testable Rule

Phase 3 taught us calculations such as SMA, RSI, MACD, ATR, Bollinger Bands, and Stochastic.

The first Phase 4 lesson introduces a new distinction: 

indicator
→ calculation

state
→ condition that may persist

event
→ transition from one state to another

rule
→ exact action tied to that event

hypothesis
→ claim that can be tested

A simple SMA crossover is used because the mathematics is already familiar. That lets us focus on research design instead of learning another indicator.

  1. How Do You Turn an Indicator into a Testable Trading Rule? From Observation to Hypothesis with Python

What You Should Understand After Rule Definition

  • An indicator is a calculation, not a trading rule.
  • A state can remain true for several bars, while an event describes a transition.
  • A testable rule must define the exact input, parameter, comparison operator, and event condition.
  • Signal time and execution time are different concepts.
  • If a signal depends on the completed Close of bar t, that information is not available before bar t closes.
  • Rule logic should be verified on artificial data before market performance is interpreted.
  • A chart can verify where events occurred, but it cannot prove predictive value.
  • The rule should be frozen before parameter hunting begins.

2. Forward Outcomes & Baselines

Once an event is defined, the next question is not immediately: 

Did the strategy make money?

A simpler question comes first: 

What happened
after the event?

Forward returns let us attach future outcomes to a signal while keeping the signal calculation itself in the past. 

signal at bar t
      ↓
future horizon
      ↓
1-bar outcome
5-bar outcome
20-bar outcome
      ↓
signal group
vs
baseline group
  1. What Is a Forward Return? Measure What Happened After a Signal with Python
  2. Is the Signal Different from the Baseline? Compare Forward-Return Distributions with Python

What You Should Understand After Forward Returns

  • A forward return measures an outcome after a reference event.
  • Close-to-future-Close return can be used as a descriptive future label.
  • If the signal is known only after bar t closes, an execution-aligned label can begin at bar t+1 Open.
  • The same forward-return definition should be applied to both the signal group and the baseline group.
  • Mean, median, positive rate, and sample count describe different properties of the outcome distribution.
  • A positive historical average does not by itself prove a predictive edge.
  • Overlapping forward-return windows can create dependent observations.
  • Comparison with ordinary market behavior is more informative than looking at signal returns alone.

3. Dependence & Uncertainty

A large table of forward returns can look convincing, but many rows may share the same future market movement. 

many rows
≠
many independent observations

Phase 4 therefore checks overlapping forward-return windows before estimating uncertainty. 

overlap
→ dependence
→ resampling design
→ bootstrap distribution
→ confidence interval
  1. Why Do Overlapping Forward Returns Matter? Understand Dependent Observations with Python
  2. How Certain Is the Result? Build Bootstrap Confidence Intervals with Python

What You Should Understand After Dependence & Uncertainty

  • Many forward-return rows do not automatically mean many independent observations.
  • Overlapping horizons can mechanically share much of the same future price movement.
  • A non-overlapping sample is a diagnostic, not proof of statistical independence.
  • A point estimate does not describe its own uncertainty.
  • A bootstrap repeatedly rebuilds a sample and recalculates the same statistic.
  • For time-series data, the resampling design matters because local dependence can be important.
  • A confidence interval is evidence under a chosen resampling design, not proof of a durable trading edge.

4. From Events to Positions

An event study still does not tell us what happens while a position remains open. 

CASH
↓
entry event
↓
BUY at next Open
↓
LONG
↓
exit event
↓
SELL at next Open
↓
CASH

The next lesson converts isolated entry and exit events into a position state that persists through time.

  1. From Events to Positions: Build a Trading Position Lifecycle with Python

What You Should Understand After Position Lifecycle

  • An event, an execution, a position, and a completed trade are different objects.
  • If a signal is known after Close(t), execution can occur at the next Open rather than retroactively at the same Close.
  • A position state can persist for many bars.
  • The current position determines whether a new event should trigger an action.
  • A lifecycle table records the position state bar by bar.
  • A trade ledger records completed Entry → Exit episodes.
  • Position lifecycle is necessary for a backtest, but it is not yet a full backtest.

Why Phase 4 Starts with Events Instead of Full Strategies

A full strategy mixes many assumptions together: 

signal
+ entry
+ exit
+ holding period
+ transaction costs
+ slippage
+ position size
+ portfolio rules
+ benchmark
+ performance metric

If all of those appear at once, it becomes difficult to understand why a result changed.

Phase 4 therefore builds the research process in layers. 

first
→ define one event

then
→ measure one outcome

then
→ compare with one baseline

then
→ quantify uncertainty

only later
→ assemble a full backtest

Signal Time Comes Before Return Calculation

One of the most important habits in trading research is to preserve chronology. 

What information existed
at decision time?

        ↓

When was the signal known?

        ↓

When could an order
actually be executed?

        ↓

Only then:

What future data
may be used as an outcome?

This prevents future information from leaking backward into the signal.

Baseline Before Edge

A signal can have a positive average return simply because the underlying market tended to rise.

So Phase 4 does not ask only: 

Was the signal return positive?

It also asks: 

What would usually have happened
without conditioning on that signal?

Possible baselines can include: 

all eligible bars

buy and hold

unconditional forward returns

randomized event timing

another simple rule

What Comes Later

The current Phase 4 path now covers rule definition, forward outcomes, baseline comparison, dependence, uncertainty, and position lifecycle. The next lessons will add the first full backtest, trading frictions, performance measurement, and robustness tests. 

Indicator Observation
→ Testable Rule
→ Forward Return
→ Baseline Comparison
→ Return Distribution
→ Confidence Interval / Bootstrap
→ Position Lifecycle
→ First Backtest
→ Transaction Costs & Slippage
→ Performance Metrics
→ Parameter Sensitivity
→ Train / Validation / Test
→ Walk-Forward Evaluation

Lessons through Position Lifecycle are now published and linked above. Later lessons should be added to this hub only after their live URLs are confirmed.

A Rule for Learning Trading Research

Do not begin by asking: 

Which indicator makes
the most money?

Begin with five smaller questions:

  1. What exactly is the observation or event?
  2. What information was available when the signal was created?
  3. When could the action realistically occur?
  4. What future outcome will be measured?
  5. What baseline will the result be compared with?

Only after those questions are fixed does it make sense to discuss statistical significance, profitability, optimization, or robustness.

From Phase 3 to Phase 4

Phase 3

raw market data
→ mathematical transformation
→ indicator
→ understand what it measures


Phase 4

indicator observation
→ precise event
→ future outcome
→ baseline
→ evidence
→ trading research

Phase 3 asks: 

What does this indicator measure?

Phase 4 asks: 

Does a precisely defined observation
contain useful information
about what happens next?

Where You Are in the Learning Path

Start Here → Market Data → Candlesticks → Indicators → Trading Research

If you are new to Phase 4, begin with the first lesson:

How Do You Turn an Indicator into a Testable Trading Rule? From Observation to Hypothesis with Python