Directional Change Trading: Event-Based Regime Signals

In the previous section, we covered the basics of Hidden Markov Models (HMM), and how they are commonly used in finance to identify market regimes such as bull, bear, and sideways. The standard approach is to fit an HMM using returns or return volatility.

However, Chen and Tsang (2021) proposed a different direction: instead of describing the market “by time,” we describe it “by events.” Specifically, they use Directional Change (DC) indicators as inputs to an HMM to detect regime transitions.

Their empirical results suggest that HMM + DC can capture regime changes more effectively than an HMM based on traditional return-volatility features. In this section, we’ll explain DC in a practical way and outline a simple trading setup to evaluate whether the equity curve improves.

What is Directional Change (DC)?

Directional Change was introduced by Guillaume et al. (1997) in the context of identifying hidden patterns in the Forex market. If you’ve used the ZigZag indicator before, DC will feel familiar—both aim to filter market noise and keep only meaningful moves.

The key difference is the question DC asks:

• Instead of: “What is the price every 5 minutes?”
• DC asks: “When does price change direction by at least X%?”

So in DC, the market is not measured in minutes or hours, but in directional-change events.

The threshold 𝜃 — the heart of DC

Everything in DC revolves around a threshold θ\thetaθ: the percentage move you consider large enough to matter.

Example:

𝜃=2%

• If the market is falling, an Upturn DC event is confirmed only when price rises at least 2% from the most recent local low.
• If the market is rising, a Downturn DC event is recorded when price falls at least 2% from the most recent local high.

Why is Directional Change useful?

(1) Noise reduction. Small fluctuations that do not reach θ\thetaθ are ignored, helping you focus on moves that actually matter.

(2) A “second perspective” on the market. Time-series data tells you how price evolves over time. DC also tells you how much time it takes for price to move by a meaningful amount. These two views complement each other and can reveal risks/opportunities that a single view might miss.

The structure of Directional Change (DC)

In the DC framework, a trend is not simply “up” or “down.” It is decomposed into two parts:

1. Directional Change event (DC) – the reversal-confirmation segment
2. Overshoot event (OS) – the continuation segment after confirmation

Together, they form the full DC-defined trend.

Extreme points (peaks and troughs)

DC divides the market into peaks and troughs, but with an important twist:
 an extreme point is confirmed only after the price has moved far enough.

• A trough is confirmed only when price rises by at least θ\thetaθ from that trough.
• A peak is confirmed only when price falls by at least θ\thetaθ from that peak.

This implies that DC extreme points are confirmed in hindsight, not immediately when they occur.

The DC event (Directional Change Event)

The DC event marks the start of a new trend.

Assume the market is in a downtrend and forms a trough at point A. Price starts to recover. When price rises by exactly the threshold θ\thetaθ and reaches point B, the segment AB is the DC event.

Point B is called the DC Confirmation Point (DCC).
At this moment, the trend is officially confirmed to have reversed from down → up.

The Overshoot event (OS)

In reality, price rarely stops right after confirmation at B. Because of momentum, it often continues moving in the same direction.

The move from B to the next extreme point C is called the Overshoot.

This overshoot segment is often the most valuable to traders, because much of the profit tends to come from the “tail” (OS) rather than the initial DC segment.

Basic quantitative measures in DC

Tsang and co-authors proposed several simple measures to quantify trend strength and behavior under DC. Common ones include:

(1) TMV – Total Price Movement

TMV measures the magnitude of a trend, normalized by 𝜃 : 

Interpretation:

• TMV=1: price barely hits the 𝜃 threshold and reverses—almost no overshoot
• TMV=3: a strong trend where overshoot dominates (a long continuation tail)

(2) T – Time

T is the physical time needed to complete a trend (trough → peak or peak → trough).

• Smaller T → faster market, higher frequency behavior
• Larger T → slower, longer-lasting trends

(3) R – Time-adjusted DC return

This can be seen as the “speed” of the market under DC:

• High R: large moves happening quickly
• Low R: weak or slow trends

A trading framework combining HMM and DC

DC is powerful, but using a fixed threshold 𝜃 (e.g., always 0.5%) has a weakness: it does not adapt to changing market conditions.

• In high-volatility periods: small 𝜃 → too many noisy signals
• In sideways markets: large 𝜃 → very few signals

To address this, a more advanced line of work (Shicheng Hu, Danping Li, and Bing Wu) proposes an Intelligent Trading Algorithm (ITA) that combines Improved DC with HMM

Step 1: Improved DC with a dynamic threshold

Instead of fixing θ\thetaθ, the strategy uses Bayesian Optimization to select the best parameters for each market phase.

They also introduce an attenuation coefficient 𝛼 for downward moves:

• Threshold for up moves: 𝜃
• Threshold for down moves: 𝛼 × 𝜃

This makes the algorithm more sensitive to drawdowns (faster exits/stop-loss behavior) while remaining more patient during upward trends.

Step 2: Regime detection using HMM

Financial markets are not homogeneous; they shift between regimes such as:

• Normal regime: stable volatility, clearer trends → easier to trade
• Abnormal regime: extreme volatility and noise → higher risk, more whipsaws (e.g., crisis/news shocks)

HMM is used to detect these shifts.

• HMM input: a DC-derived return series such as RDC​. Since DC captures meaningful movements and filters noise, this input can be “cleaner” than traditional volatility-based features.
• HMM output: a regime label, e.g., Regime 1 (Normal / safer) vs Regime 2 (Abnormal / riskier).

Step 3: A unified trading rule set

Think of the final strategy as an intelligent filter:

Entry:

• When an Upturn DC event is confirmed (price rises beyond θ\thetaθ), check the HMM:
  • If HMM indicates Normal regime → BUY
  • If HMM indicates Abnormal regime → STAY OUT

Exit:

• Take profit: when the overshoot reaches a target based on empirical regularities (e.g., an additional move of about 𝜃)
• Stop loss: when a Downturn DC event occurs (price drops beyond 𝛼 × 𝜃)

The example below illustrates the full pipeline from DC to HMM to portfolio allocation. We download BTC and XRP data from yfinance, compute Directional Change features (notably RRR), and feed them into an HMM to infer market regimes. Using the inferred regime and DC features, we optimize BTC/XRP portfolio weights and backtest by applying those weights to realized returns to build an equity curve. Finally, we compare the strategy to Buy & Hold using final equity, Sharpe ratio, and maximum drawdown.

Conclusion

Directional Change (DC) offers an event-based view of price action that filters out small, noisy fluctuations and focuses on moves that actually matter. When DC-derived features are used as inputs to an HMM, regime shifts can be identified more cleanly than with traditional time-based volatility or returns alone, which helps avoid trading during unstable periods. In practice, combining DC + HMM provides a simple but effective framework: detect the market’s current regime, adapt risk exposure accordingly, and concentrate on the overshoot portion where trends often deliver most of the payoff. While performance will depend on the choice of threshold and optimization settings, this approach provides a robust foundation for building adaptive, regime-aware trading strategies.