OnlineTechnicalIndicators internals

Sub-indicator(s)

An indicator can be composed internally of sub-indicator(s). Input values catched by fit! calls are transmitted to each sub_indicators to be processed to _calculate_new_value function which calculates value of indicator output.

Example: Bollinger Bands (BB) indicator owns 2 internal sub-indicators

• central_band which is a simple moving average of prices,
• std_dev which is standard deviation of prices.

Composing new indicators

Indicators Chaining with OnlineStatsChains

Indicators can be chained together to create complex compositions - it's like building new indicators with Lego bricks.

Examples of chained indicators:

• DEMA : 2 moving averages chained together
• TEMA : 3 moving averages chained together
• T3 : 6 moving averages chained together
• TRIX : 3 moving averages chained together with ROC calculation

Using OnlineStatsChains StatDAG (Recommended)

The recommended way to compose indicators is using OnlineStatsChains.StatDAG. This provides a directed acyclic graph (DAG) structure for organizing indicator chains with automatic value propagation through filtered edges.

Basic pattern:

using OnlineStatsChains

# 1. Create a StatDAG
dag = StatDAG()

# 2. Add indicator nodes with symbolic names
add_node!(dag, :ema1, EMA(period=10))
add_node!(dag, :ema2, EMA(period=10))

# 3. Connect nodes with filtered edges (only propagates non-missing values)
connect!(dag, :ema1, :ema2, filter = !ismissing)

# 4. Fit data to the source node - values propagate automatically
fit!(dag, :ema1 => 100.0)

# 5. Read values from any node
value(dag, :ema1)  # First EMA value
value(dag, :ema2)  # Second EMA value (automatically updated)

Benefits of StatDAG:

• Clear structure: Visual organization of the indicator pipeline
• Automatic propagation: Values flow through filtered edges without manual handling
• Named access: Each stage can be inspected by name for debugging
• Flexible composition: Easy to add/remove stages or modify connections

DAGWrapper: Integration with fit! Infrastructure

The DAGWrapper struct provides a compatibility layer between StatDAG and the OnlineTechnicalIndicators' fit! infrastructure:

mutable struct DAGWrapper
    dag::StatDAG        # The underlying StatDAG
    source_node::Symbol # Entry point for data
    stats::Vector{OnlineStat}  # For compatibility checks
end

When fit! is called on a DAGWrapper, it forwards the data to the DAG's source node, which then automatically propagates values through all connected edges:

function fit!(wrapper::DAGWrapper, data)
    fit!(wrapper.dag, wrapper.source_node => data)
end

This allows composed indicators to use StatDAG internally while maintaining compatibility with the standard fit! interface.

Example: How DEMA Uses StatDAG

The DEMA (Double Exponential Moving Average) indicator demonstrates the StatDAG pattern:

Structure:

Input → :ma1 (EMA) → :ma2 (EMA) → DEMA calculation

Implementation:

# In the DEMA constructor:
dag = StatDAG()
add_node!(dag, :ma1, MAFactory(Float64)(EMA, period=period))
add_node!(dag, :ma2, MAFactory(Float64)(EMA, period=period))

# Filtered edge - only propagate non-missing values
connect!(dag, :ma1, :ma2, filter = !ismissing)

# Wrap for compatibility with fit! infrastructure
sub_indicators = DAGWrapper(dag, :ma1, [dag.nodes[:ma1].stat])

How it works:

1. When fit!(dema, data) is called, data flows to sub_indicators (the DAGWrapper)
2. DAGWrapper forwards to dag[:ma1]
3. The filtered edge automatically propagates to dag[:ma2] (if value is non-missing)
4. In _calculate_new_value, values are read from both nodes:

function _calculate_new_value(ind::DEMA)
    val2 = value(ind.dag, :ma2)
    if !ismissing(val2)
        # DEMA formula: 2 * MA1 - MA2
        return 2.0 * value(ind.dag, :ma1) - val2
    else
        return missing  # Chain not fully warmed up
    end
end

Key insight: The filtered edge (filter = !ismissing) eliminates the need for nested conditionals - propagation only happens when values are available.

For migrating from the old add_input_indicator! API, see the Migration Guide.

Filtering and Transforming Input (Legacy)

The legacy mechanism allows filtering and transforming input of an indicator using functions (typically anonymous functions). Input can be filtered/transformed before being passed to sub-indicators or processed by _calculate_new_value.

This mechanism still exists in the codebase for backward compatibility with custom indicators that may have implemented input_filter and input_modifier fields, but is no longer used by any built-in indicators.

Moving average factory

• SMA, EMA, ... are moving average.

Most complex indicators uses in their original form SMA or EMA as default moving average.

In some markets they can perform better by using instead an other kind of moving average.

A moving average factory have been implemented

This kind of indicators have a ma parameter in order to bypass their default moving average uses.