Advanced Filtering Systems

1. Overview

Advanced filtering systems are essential tools in technical analysis for determining precise entry and exit points. When a trading signal is generated, the process of distinguishing genuine signals from false ones is called filtering—and the quality of these filters ultimately determines the overall profitability of a trading system.

This chapter systematically classifies the three major filtering methods: Price-Based, Time-Based, and Event-Based. It examines the characteristics and limitations of each filter type. Additionally, it addresses the variable stop-loss size problem that inevitably arises with breakout entries, and introduces Proportional Stop-Sizing as the solution. This method dynamically calculates position size in response to the varying stop-loss distance of each trade, enabling consistent risk management across all trades.

2. Core Rules and Principles

2.1 Three Categories of Filtering Systems

All filtering systems can be classified into three broad categories, each providing different types of information—and each failing to provide other types. Understanding this distinction is the starting point for constructing effective filter combinations.

Price-Based Filters

• Characteristic: Provide an exact entry price, but do not specify when that price will be reached.
• Key Advantage: The distance to the stop-loss can be calculated precisely before entry, making risk control superior to other filter types.
• Sub-classifications:
  • Absolute Measure: Entry requires price to exceed a fixed amount beyond the breakout level (e.g., breakout line + 10 points)
  • Relative Measure: Entry requires price to exceed a fixed percentage beyond the breakout level (e.g., breakout line + 0.5%)
  • Volatility Measure: Entry requires price to exceed a multiple of ATR (e.g., breakout line + 1.0 ATR) or a multiple of standard deviation

Practical Tip: In highly volatile markets like cryptocurrency, volatility-based measures (ATR-based) are more adaptive than absolute measures. The filter width automatically adjusts according to current market conditions.

Time-Based Filters

• Characteristic: Provide an exact entry time, but do not specify the entry price at that time.
• Sub-classification: Entry is confirmed only after price remains above (or below) the breakout level for N consecutive bars (candles).
• Limitation: Since it is unknown how far price will have moved during those N bars, the entry price at the time of confirmation is unpredictable. This means the stop-loss size cannot be determined in advance.

Example: A classic time-based filter requires three consecutive closing prices above a resistance level before confirming entry.

Event-Based Filters

• Characteristic: Neither the exact price nor the exact timing is known until a specific event occurs.
• Sub-classifications:
  • Algorithmic Filters: Entry requires fulfillment of predefined rules such as specific bar sequences, or the order in which new highs/lows occur
  • Event-Based Measures: Entry is triggered by events such as a close violation (closing price breaching a specific level) or a re-test of the broken barrier after a confirmed breakout

Practical Example: After price breaks above the upper boundary of a triangle pattern, waiting for price to pull back to the breakout level, confirm it as support (pullback re-test), and then entering—this is a classic application of an event-based filter.

2.2 Dual Filtering Systems

A single filter alone cannot simultaneously achieve both risk control and signal precision. Time-based and event-based filters in particular cannot specify the entry price in advance, so a dual filtering system that combines a price-based filter as a secondary layer is necessary.

Application Conditions:

• When the maximum allowable risk per trade must be limited and controlled
• When using time-based or event-based filters, a price-based filter is added as a secondary layer for risk control

Example Application:

• Primary Filter: Uses a Close Violation as the entry trigger → a signal is generated when the closing price settles beyond the breakout level
• Secondary Filter: A price-based filter restricts entry to within a specific distance from the breakout level → if the closing price has moved too far from the breakout level (e.g., exceeding 2 ATR), the signal is ignored to prevent excessive risk exposure

The core principle of dual filtering is the division of roles: the primary filter governs signal quality, while the secondary filter governs risk limits. When the secondary filter rejects a signal, the trade is skipped entirely—the filter conditions are never relaxed.

2.3 Proportional Stop-Sizing Procedure

The inherent problem with breakout entries is that the distance to the stop-loss varies with every trade. Barrier entries place the stop just behind the support/resistance level, resulting in a nearly constant stop-loss size. Breakout entries, however, measure from the breakout point to the most recent significant swing high/low, and this distance changes every time.

The Proportional Stop-Sizing method solves this problem through the following 5-step procedure:

Step	Procedure	Description
1	Run Backtest	Simulate a minimum of 300–500 trades using the strategy and record the stop-loss size (distance from entry to stop) for each trade.
2	Calculate 2-Standard-Deviation Value	Compute the standard deviation of all collected stop-loss size samples, then multiply by 2.
3	Determine Proportional Stop Size	Mean stop-loss size + (2 × standard deviation) = Proportional stop size. This value serves as an upper bound that encompasses approximately 95% of all trades.
4	Set Maximum Risk Percentage	Define the maximum risk per trade as a percentage of current capital (e.g., 2%) and calculate the corresponding dollar (or USDT) value.
5	Calculate Proportional Position Size	Dollar risk per trade ÷ Proportional stop size = Proportional position size (base position size)

Position Sizing Rules:

Condition	Application
Actual stop size ≤ Proportional stop size	Use the proportional position size (base position size) as-is. The narrower the stop, the smaller the actual dollar risk, creating a natural safety margin.
Actual stop size > Proportional stop size	Recalculate position size as maximum dollar risk ÷ actual stop size. This ensures the maximum risk limit per trade is never exceeded.

Core Principle: This method maintains position size on trades with narrow stops to capitalize on favorable risk/reward ratios, while reducing position size on trades with wide stops to keep risk within limits. The result is lower equity curve volatility and reduced drawdowns compared to a fixed-ratio approach.

3. Chart Verification Methods

3.1 Verifying Filter Effectiveness

After introducing a filter, quantitative verification is mandatory. The key is to confirm that the filter reduces false positive signals without excessively filtering out valid ones.

Price-Based Filter Verification:

• Use backtesting to compare performance without the filter side by side with performance after applying the filter.
• Measure the false breakout signal reduction rate—what percentage of false breakouts did the filter eliminate?
• Check the consistency of entry prices. Determine whether the deviation of entry prices relative to the breakout level has decreased.
• Caution: If the filter width is too large, the entry price moves further from favorable levels, potentially degrading the risk/reward ratio.

Time-Based Filter Verification:

• Measure the win rate and expected return of signals that met the N-bar persistence condition before entry.
• Compare the total performance of early entry (no filter) versus delayed entry (filter applied). Even if the win rate improves, the entry price may become less favorable, potentially reducing expected returns. Therefore, both win rate and profit factor must be evaluated simultaneously.

Event-Based Filter Verification:

• Measure the success rate of trades entered after a specific event (pullback re-test, close violation, etc.) occurs.
• Analyze performance differences compared to trades entered without the event occurring.
• Event-based filters can significantly reduce trade frequency, so verify that a sufficient sample size is available to ensure statistical significance.

3.2 Proportional Stop-Sizing Verification

Barrier Entry vs. Breakout Entry Comparison:

Factor	Barrier Entry	Breakout Entry
Stop Placement	Just behind the barrier (support/resistance)	Behind the most recent significant swing high/low
Stop Size	Nearly constant	Varies with each trade
Position Sizing	Fixed-ratio method applicable	Proportional sizing required
Entry Location	Near the barrier (favorable price)	After breakout confirmation (less favorable price)

Stop-Loss Size Distribution Analysis:

• Calculate the mean and standard deviation of stop-loss sizes from a minimum of 300–500 trade samples.
• Generate a histogram of stop-loss sizes to examine the distribution pattern. The closer to a normal distribution, the more stable and effective the proportional stop-sizing method will be.
• If extremely wide stops (3+ standard deviations) occur frequently, the entry logic of the strategy itself should be re-examined.

4. Common Mistakes and Pitfalls

4.1 Filtering-Related Mistakes

Over-Filtering:

• Stacking too many filters drastically reduces trade opportunities, causing statistical significance to disappear. If fewer than 300 backtest trades can be generated, the system's reliability cannot be assessed.
• Complex filter combinations carry a high risk of curve fitting. Filters perfectly fitted to historical data tend to break down in live trading.

Inappropriate Filter Combinations:

• Using time-based or event-based filters alone makes it impossible to know the entry price in advance, preventing effective per-trade risk control. A price-based secondary filter must always be combined.
• Using event-based entries such as close violations without a price-based filter can result in abnormally large stop-loss sizes when entry occurs on strong momentum bars far from the breakout level.

Confusing the Purpose of Filters:

• Filters are tools to improve signal quality, not tools to generate signals. If the underlying entry signal is not robust, no amount of sophisticated filtering will improve profitability.

4.2 Proportional Stop-Sizing Mistakes

The Fixed-Ratio Risk Management Trap:

• If every trade risks the same percentage (e.g., 2% of capital) but position size remains fixed, when narrow stops are frequently triggered, small losses accumulate rapidly.
• Conversely, when wide stops are frequently triggered, a single stop-out drains a disproportionately large amount of capital.
• Proportional stop-sizing mitigates both extremes, stabilizing the equity curve.

Insufficient Sample Size:

• Calculating the mean stop-loss size from fewer than 300 trades results in low statistical reliability, distorting the proportional stop size.
• If market conditions have changed significantly but calculations rely solely on historical data, current volatility is not reflected. Periodic updates using a rolling window (based on the most recent N trades) are necessary.

Incorrect Proportional Threshold Settings:

• Using 1 standard deviation covers only approximately 68% of trades, causing frequent position size adjustments on the remaining 32%. This undermines system consistency.
• Using 3 standard deviations covers nearly all trades, but the proportional stop size becomes so large that the base position size shrinks excessively.
• 2 standard deviations (approximately 95% coverage) represents the optimal balance between consistency and efficiency.

5. Practical Application Tips

5.1 Filter Selection Guidelines

Fundamental Principles:

• Prioritize price-based filters. They allow the stop-loss size to be determined before entry, providing the clearest risk management.
• When using time-based or event-based filters, always add a price-based secondary filter to cap maximum risk.
• Always design filters with backtestability in mind. Avoid subjective filters that cannot be clearly defined as rules.

The Advantage of Volatility-Based Filters:

• Cryptocurrency markets experience extreme volatility swings, making ATR-based volatility filters far more adaptive than fixed-price or fixed-percentage filters.
• Example: Using 0.5× the 20-period ATR as a breakout filter causes the filter to narrow during low-volatility periods and widen automatically during high-volatility periods.

Combining with Trend Filters:

• The MACD signal line crossover can serve as a trend direction filter.
• Long Entry Condition: Stochastic crosses above its signal line + MACD is positioned above the zero line → uptrend confirmed
• Short Entry Condition: Stochastic crosses below its signal line + MACD is positioned below the zero line → downtrend confirmed
• Combining an oscillator (timing) + trend indicator (direction) in this way effectively filters out counter-trend entries.

5.2 Practical Implementation Strategy

Staged Filter Application:

1. Stage 1 — Direction Confirmation: Confirm the major trend direction using moving average crossovers, MACD direction, or whether price is above/below the 200 EMA. Ignore signals that oppose this direction.
2. Stage 2 — Price-Based Primary Filter: Refine the entry point using the breakout level + ATR multiple, percentage threshold, etc. The stop-loss size is determined at this stage.
3. Stage 3 — Time/Event-Based Secondary Filter: If needed, enhance signal precision through N-bar persistence confirmation, pullback re-test, etc. However, even after applying the secondary filter, the price limits established in Stage 2 must not be exceeded.

Proportional Stop-Sizing in Practice:

• Regular Updates: Re-run backtests at least quarterly to recalculate the mean stop-loss size and standard deviation. Update more frequently when market structure changes (e.g., before and after Bitcoin halving events).
• Automation: Prepare a spreadsheet or script that measures the stop-loss size before each trade, compares it against the proportional position size, and automatically calculates the appropriate position size.
• Exception Handling: In extreme cases where the actual stop-loss size exceeds 2× the proportional stop size, even reducing position size may result in an unacceptably poor risk/reward ratio. It is advisable to add a rule to skip such trades entirely.

Integration with Risk Management:

• Harmonize the proportional sizing method with account-level maximum risk limits (e.g., total open position exposure capped at 6% of capital).
• Link it with rules for additional position size reduction during consecutive losses (e.g., reduce position size by 50% after 3 consecutive losses).
• Maintain consistency between the filtering system and drawdown limit conditions (e.g., halt trading when maximum drawdown reaches 15%). Resist the temptation to relax filters in order to increase trade frequency.

5.3 Performance Monitoring

Key Metrics to Track:

Metric	Measurement Method	Significance
Win Rate Change	Compare before and after filter application	Effectiveness of false signal elimination
Average Win/Loss Ratio	Average winning trade ÷ Average losing trade	Impact of filtering on risk/reward
Profit Factor	Gross profit ÷ Gross loss	Overall system profitability
Maximum Drawdown	Largest peak-to-trough decline in equity	Risk management effectiveness
Trade Frequency	Number of trades per month/quarter	Confirms filters are not excessively reducing opportunities

Optimization Precautions:

• Verify filter performance individually across different market conditions (trending, ranging, high-volatility, low-volatility). Filters that work only under specific conditions lack robustness.
• To prevent over-optimization, always conduct out-of-sample testing. The standard approach is to optimize on 70% of the data and validate on the remaining 30%.
• In live trading, slippage and execution delays occur—unlike in backtesting. Breakout entries are particularly susceptible since orders concentrate at moments of low liquidity. Apply a conservative 0.1–0.3% slippage per trade to backtest results for realistic evaluation.
• When fine-tuning filter parameters, if performance reacts sensitively to parameter changes, this is a sign of over-optimization. Select settings that demonstrate stable performance across a wide parameter range.