Languages: English | Ελληνικά

Dedicated to long-term investors:

We don’t panic sell... we DCA the dips so hard they file a class‑action restraining order.

DCAi is a machine learning-driven Dollar-Cost Averaging framework that combines K-Nearest Neighbors pattern recognition with a dynamic three-tier decision engine (Pullback, Oversold, Fear).

It implements a Savings Pot (budget carryover) and exponential position scaling (Rho), with entries gated by probabilistic pattern matching and confidence thresholds.

The strategy auto-calibrates for four asset classes (Crypto, Stocks, Indices, Commodities), adjusting entry thresholds and position sizing to match each market's volatility characteristics.

1. The Inefficiency of Static DCA

   • Problem 1: Wasted Capital During Sideways/Bull Markets
   • Problem 2: Uniform Position Sizing Ignores Volatility
   • Problem 3: Capital Exhaustion in Prolonged Bear Markets

2. Core Architecture

   • 2.1 Machine Learning Engine (KNN)
     • 2.1.1 How It Works
     • 2.1.2 Technical Summary
   • 2.2 Adaptive Asset Sensitivity
     • 2.2.1 The Rho (ρ) Sensitivity Parameter
     • 2.2.2 Adaptive Thresholds by Asset Class
   • 2.3 Quick Reference

3. Decision Engine Logic

4. Financial Engineering & Budgeting

   • 4.1 Smart Budgeting (The Savings Pot)
   • 4.2 Dynamic Position Sizing

5. Comparative Analysis: Pros & Cons

   • 5.1 Advantages of DCAi
   • 5.2 Limitations & Risks

6. Future Research & Roadmap

7. Configuration & Parameters

   • 7.1 Asset Selection & Auto-Optimization
   • 7.2 Machine Learning Settings (KNN)
   • 7.3 Financial Parameters (Budgeting)
   • 7.4 Technical Confirmation (Filtering)
   • 7.5 Recommended Configurations
   • 7.6 Parameter Tuning & Optimization Guide
   • 7.7 Parameter Interactions & Dependencies
   • 7.8 Community Settings & Contributions

8. Frequently Asked Questions (FAQ)

9. References & Academic Foundation

10. Disclaimer & Risk Warning

Traditional Dollar-Cost Averaging (DCA) is a passive strategy that executes purchases at fixed intervals regardless of market valuations. While psychologically effective for retail investors, it suffers from several mathematical inefficiencies:

Issue: Static DCA allocates the full monthly budget every period, regardless of market regime. During bull markets or sideways consolidation, this means deploying capital at progressively higher prices and missing opportunities to allocate more during large drawdowns.

Why This Matters: If price drops sharply after monthly capital is already deployed, fewer units are accumulated at lower prices.

DCAi Mechanism (see Section 4.1: Savings Pot): Instead of forced monthly deployment, DCAi accumulates unused capital into a Savings Pot. When the ML engine detects a reversal signal (Section 3: Decision Engine Logic), it can deploy this reserve alongside the base budget.

Issue: Static DCA invests the same dollar amount every month, regardless of whether the asset is trading near all-time lows or all-time highs. This creates an inverse relationship between expected return and capital deployment—you buy more when valuations are worst... except in terms of Capital Exposure. While Static DCA buys more units mathematically, it fails to increase Fiat Capital allocation during deep discounts, treating a -5% dip and a -50% crash with the same financial urgency.

Why This Matters: Asset volatility varies dramatically across markets. Static DCA treats different volatility regimes identically.

DCAi Mechanism (see Section 4.2: Dynamic Position Sizing and Section 2.2: Adaptive Asset Sensitivity): DCAi uses Inverse-Price Weighting with an exponential $\rho$ (Rho) parameter. As price falls relative to the historical average, buy sizes increase exponentially. The $\rho$ value is automatically adjusted per asset class to match volatility signatures (Crypto: 1.7, Stocks: 2.0, Indices: 2.5).

Issue: Static DCA commits all capital upfront on a predictable schedule. In prolonged bear markets, this can leave less capital available for later drawdowns.

Why This Matters: When large drawdowns arrive late in a cycle, fixed schedules may leave fewer funds available to allocate at those levels.

DCAi Mechanism (see Section 4.1: Savings Pot and Section 3: Decision Engine Logic): The Savings Pot preserves capital during non-signal periods. The ML engine's three-tier decision framework (Tier 1 Pullback < Tier 2 Oversold < Tier 3 Fear) uses probabilistic gating to reduce entries during low-conviction periods and concentrate capital during higher-conviction reversals. Combined with (see Section 7.3: Pot Reserve %), a percentage of the pot can be reserved for extreme-case deployments.

DCAi searches historical price data to find past situations that match current market conditions. When the market looks oversold today, the algorithm scans back through the configured lookback window to find the K most similar moments. It then checks what happened next—did price rally or keep falling?

The K parameter controls how many historical examples to examine. K=5 means "find the 5 most similar past situations." If 4 out of 5 led to price increases, that's 80% confidence. Higher K values require more historical agreement before triggering a buy, making the system more conservative.

The Three Market Conditions (Features):

DCAi compares three metrics to identify similar market states:

1. Money Flow Index (MFI) — Measures buying vs. selling pressure on a 0-100 scale. Values below 20 indicate panic selling, above 80 suggests euphoric buying. DCAi targets oversold zones between 0-55 depending on the asset.

2. Rate of Change (ROC) — Tracks price velocity. Negative values mean falling prices; the more negative, the steeper the decline. This filters for actual dips rather than sideways drift.

3. Average True Range (ATR) — Quantifies volatility magnitude. High ATR during selloffs signals market distress, which often precedes reversals.

Lorentzian Distance:

Standard distance metrics (Euclidean) treat a 50% crash as fundamentally different from a 40% crash. Lorentzian Distance uses logarithmic scaling to match patterns by shape and direction rather than absolute magnitude.

• Model: K-Nearest Neighbors (KNN) non-parametric classifier
• Features: MFI, ROC, ATR normalized to percentile ranks
• Distance Metric: Lorentzian Distance (log-damped)
• Prediction Window: 4-bar forward-looking
• Confidence: ML probability (0-100%) gates entries and scales position size
• Thresholds: 70% minimum for oversold entries, 50% for pullback entries

Rho determines how much your position size increases as price falls below its historical average. Static DCA deploys the same dollar amount every period. DCAi scales exponentially—the deeper the dip, the larger the buy.

With a €50 monthly budget on Bitcoin:

Lower Rho (1.5-1.7) suits volatile assets where -40% drawdowns happen frequently. Higher Rho (2.0-2.5) suits stable assets where deep dips are rarer and position sizing can scale more during those periods.

Crypto can stay oversold for months—a moderate Rho can reduce premature capital exhaustion. Index crashes (2008, 2020) are rare events where higher Rho implies larger sizing during deep drawdowns.

MFI thresholds vary by asset because "oversold" is relative to market structure:

When you select an asset class, DCAi calibrates both the oversold threshold and position scaling to match that market's typical behavior.

DCAi compares current market conditions (MFI, ROC, ATR) against historical data to find similar patterns, checks whether those patterns were followed by short-term price increases, then scales position size based on dip severity and pattern confidence.

The strategy prioritizes trades into three distinct tiers based on signal conviction and liquidity availability:

1. Tier 1: PULLBACK BUY (Trend Following)
   • Occurs in healthy uptrends when price trades in the "discount zone" below the Ichimoku Kijun-sen, above Leading Span B, and the cloud is green.
2. Tier 2: OVERSOLD BUY (Standard Dip)
   • Activated in oversold conditions (MFI < 35), with dynamic relaxation when CVD divergence is bullish, and ML confirmation.
   • Uses a Strong Boost multiplier (default 1.5x).
3. Tier 3: FEAR BUY (Extreme Panic)
   • Activated when MFI falls below the "Panic" threshold (default 20), with dynamic relaxation when CVD divergence is bullish.
   • Uses a Max Multiplier and aggressive pot allocation.

If no buy signal is triggered within a calendar month, the monthly budget is automatically rolled over into a Savings Pot. This "carryover" mechanism allows larger capital deployment during extreme market lows.

The investment amount is calculated using an Inverse-Price Weighting formula:

• Exponential Scaling: As price falls relative to the historical average, the buy amount increases exponentially based on the $\rho$ parameter.
• Confidence Boost: ML probability acts as a multiplier; higher confidence increases the multiplier and pot usage on a continuous scale (starting above 50% confidence).

• Dynamic Sensitivity: Utilizes the $\rho$ parameter to exponentially increase buy size during large deviations.
• ML-Driven Confirmation: The KNN engine requires historical pattern similarity before signaling a buy.
• Capital Preservation: The Savings Pot preserves capital for periods when signals occur.

• Computational Limits: KNN models in Pine Script are limited by the lookback window (max 3000 bars).
• Overfitting Risk: Lorentzian distance parameters may overfit to recent price action.
• Execution Latency: Relies on bar closes, which may result in higher entries during fast recoveries.

1. Sentiment Integration: Incorporating on-chain data or Funding Rates as additional features.
2. Multi-Timeframe Validation: Correlating 4-bar predictions across Daily and Weekly intervals.
3. Recursive Learning: Implementing a self-correcting mechanism for the $\rho$ parameter based on realized drawdown.

DCAi offers a configurable settings menu to align the algorithm with your risk profile and asset class.

• Asset Class: Choose between Crypto, Stocks, Indices, or Commodities. This selection automatically adjusts:
  • MFI Thresholds: Tailored to the typical volatility of each sector.
  • Adaptive Sensitivity ($\rho$): Controls how much the position size increases during dips.
• Auto-Optimize Parameters: When enabled, the script ignores manual overrides and uses pre-set values for the selected asset.
• Manual Sensitivity ($\rho$): Overrides auto-optimization when auto is disabled.

• Lookback Window: Number of historical bars (up to 2800) the ML model uses to find similar patterns.
• K-Neighbors: The number of "nearest neighbors" compared (default is 5).
• Probability Threshold: The minimum ML confidence required to trigger an entry signal (defaults to 70% for strong signals and 50% for pullback entries).
• ML Confidence Sensitivity: Controls how strongly the distance-based confidence multiplier scales position sizing.
• ROC Length (Feature): Length used for the ROC feature in the KNN input set.

• Monthly Budget: Your total investable capital per month.
• Start Date / End Date: Limits signals and budgeting to a defined trading window.
• Max Multiplier Cap: Limits the maximum investment size for a single signal to prevent over-exposure.
• Strong Buy Boost: Multiplier applied to oversold signals.
• Max Buy Boost: Multiplier applied to fear signals.
• Pot Reserve (%): Portion of the savings pot held back for larger dips.
• Show Savings Pot Usage: Displays pot usage labels on the chart when enabled.

• Ichimoku Display Options: Toggles for Tenkan/Kijun lines, Chikou span, and Kumo fill.
• Cooldown Period: Number of bars to wait between strong signals to avoid signal clustering.

Below are three pre-built profiles for backtesting, aligned with different risk tolerances and market conditions (see Section 2.1 ML settings and Section 3 decision tiers for context):

Typical use cases: Risk-averse investors, volatile assets (micro-caps), bear markets

Expected Behavior: 1-3 signals/month, deeper drawdown reserve, lower average entry prices vs. higher-frequency profiles

Typical use cases: Long-term investors, established assets (BTC/ETH, Index funds), 2-8 year horizons

Expected Behavior: 3-6 signals/month, moderate pot accumulation, balanced entries across tiers

Typical use cases: Active investors, stable assets (large-cap stocks, top-tier crypto), bull/accumulation phases

Expected Behavior: 8-15+ signals/month, faster pot depletion/replenishment cycles, earlier entries into dips

Fine-tuning DCAi for your specific asset and market regime requires systematic testing. Use this guide to iterate from a base profile:

1. Start with Profile B (Balanced) for your asset class
2. Run the indicator live or backtest for 1-3 months minimum (Start Date → End Date in Section 7.3)
   • Profile A/B typically generate 1-6 signals/month—shorter periods lack statistical significance
   • 1-3 months provides 3-18 signals for proper evaluation
3. Record: number of signals, average entry price, portfolio change, drawdown, Sortino ratio

Metric to Track: Signal Frequency = signals/month (typical range: 3-8 for balanced)

Action: Adjust one parameter per week and re-test at least 2 weeks of data before the next change.

If you want more capital at risk during dips:

• Increase Monthly Budget (larger monthly budget → bigger entries)
• Increase Strong Buy Boost and Max Buy Boost (Section 3 tiers 2-3)
• Decrease Pot Reserve % (deploy more reserve capital)
• Decrease ML Confidence Sensitivity (even lower-confidence signals get sized larger)

If you want less capital at risk (capital preservation):

• Decrease Monthly Budget
• Decrease Strong Buy Boost and Max Buy Boost
• Increase Pot Reserve % (10-30%)
• Increase ML Confidence Sensitivity (only higher-confidence signals get sized larger)
• Increase Probability Threshold (gate more entries)

Crypto (BTC/ETH): Auto Rho = 1.7 (Section 2.2)

• Crypto can stay oversold for weeks; increase Probability Threshold to 75% to filter noise
• Larger daily swings may justify a higher Monthly Budget (€100+)
• Use Profile C settings or tune conservatively within Profile B

Stocks (Tech, Blue-Chip): Auto Rho = 2.0

• Stocks recover faster; lower Cooldown Period (5-8 bars) enables faster re-entry
• More efficient markets = tighter Lookback Window (1500 bars preferred)
• Profile B defaults work well; lower K-Neighbors to 3-4 for responsiveness

Indices (S&P500, World ETFs): Auto Rho = 2.5

• Broad indices are smoother; shorter Lookback Window (1000-1500) reduces lag
• Lower signal frequency natural; Probability Threshold 65-70%
• Profile A or low-end Profile B recommended

Commodities (Gold, Metals): Auto Rho = 2.5

• Cyclical and mean-reverting; increase Lookback Window to 2500+ to capture longer cycles
• Wider trading ranges = higher Monthly Budget justified
• Profile B/C hybrid: moderate budget (€75-175) + high Rho sensitivity

1. Extended Backtest (TradingView Pine Editor): Set Start Date 6-12 months ago (minimum 1 full market cycle), run indicator

   • Generate 12-72+ signals for statistically robust comparison
   • Compare DCAi portfolio (avg entry, ROI, Sortino) vs. monthly passive DCA benchmark
   • DCAi aims for a lower average entry price and similar or higher returns relative to the passive benchmark
   • Record: How many Tier 1 (pullback) vs. Tier 2/3 (oversold/fear) signals? Are they clustered by market regime?

2. Live Paper Trading (2-4 weeks minimum): Run on live chart without capital

   • Let at least 1-2 actual signals fire during paper trading (not backtested)
   • Review how signals align with recent price structure and your thesis
   • Check Ichimoku clouds (Tier 1 Pullback validation, Section 3)
   • Observe how ML confidence (%) behaves into dips vs. recoveries (Section 2.1)
   • Validate portfolio tracking and Savings Pot accumulation logic

3. Full-Size Live (after at least 4 weeks confidence, post-paper): Deploy real capital

   • Start with smallest Monthly Budget tier (€25-50)
   • Run for at least 2-3 full months (to capture signal variance)
   • Scale up only if metrics align: lower avg entry, good entry quality, Sortino ratio positive

• Average entry price is higher than blind DCA: ML settings too loose, increase Probability Threshold by 10%+
• Portfolio never catches dips (<2 Tier 2/3 entries/year in volatile market): Decrease Probability Threshold, increase K-Neighbors, or double Monthly Budget
• Drawdown exceeds 50% through no fault of market: Max Buy Boost too aggressive; reduce to 1.5-2.0x, or increase Pot Reserve %
• Pot grows to 10x+ monthly budget and never deploys: Increasing Probability Threshold is likely counterproductive; decrease it to trigger more deployments

These parameters work together. Changing one affects optimal values of others:

We want to hear from you! If you've optimized DCAi for your specific use case and want to share your configuration, please post it in the GitHub Discussions section.

When posting your settings, please include the following context to help others evaluate whether your configuration matches their needs:

• Asset & Ticker: What are you trading? (e.g., BTC/USDT, SPY, Gold futures)
• Timeframe: Daily (D), Weekly (W), or other?
• Market Regime: Bull market, bear market, accumulation phase, or full cycle test?
• Profile Base: Which recommended profile (A/B/C) did you start from?
• Modified Parameters: List the key settings you changed from the base profile
• Performance Metrics (optional but valuable):
  • Average entry price vs. blind DCA benchmark
  • Signal frequency (signals/month)
  • Pot accumulation behavior
  • Drawdown management
  • Time period tested (backtested or live)

• Real-world validation: Community testing covers market conditions no single backtest can simulate
• Asset-specific optimization: Users trading similar assets benefit from your discoveries
• Edge cases: Unusual assets (high-vol small-caps, seasonal commodities) may need tuning beyond default profiles
• Collective refinement: Useful configurations emerge through collaboration

1. Navigate to GitHub Discussions → Share Your Settings
2. Create a new discussion post with a descriptive title (e.g., "BTC Daily - Conservative Bear Market Profile")
3. Use the template format above to provide context
4. Optional: Include screenshots of your backtest results or key signals

Note: All shared configurations are community-contributed and not officially endorsed. Always backtest and paper trade any community settings before deploying real capital (see Section 7.6, Step 5 for validation methodology).

A: Unused monthly budget accumulates into the Savings Pot. If your monthly budget is €50 but the market is rallying with no buy signals, that €50 rolls over. During the next qualifying dip, DCAi can deploy both the current month's budget plus accumulated reserves, significantly lowering your average entry.

A: Euclidean distance amplifies outliers. A 50% crash and a 40% crash would be treated as fundamentally different patterns despite both representing severe capitulation. Lorentzian Distance uses logarithmic scaling to match patterns by structural shape rather than absolute magnitude, making it more robust for financial data with frequent spikes and crashes.

A: DCAi is designed for Swing Traders and Long-term Investors. It is intended for Daily (D) or Weekly (W) timeframes. Using it on low timeframes (e.g., 1-minute or 5-minute) can result in excessive signals caused by market noise, leading to quicker capital deployment.

A: The $\rho$ (Rho) parameter determines how aggressively the algorithm scales its position size relative to price drops.

• High $\rho$: The buy amount increases exponentially as the price falls further below the mean.
• Low $\rho$: The allocation is more linear, behaving closer to traditional, flat-rate DCA.

A: Different assets have different "volatility signatures." For example, the Money Flow Index (MFI) on Bitcoin can stay oversold much longer than on the S&P 500. By selecting the correct Asset Class, the algorithm automatically re-calibrates its "Panic" and "Oversold" thresholds to match that specific market's behavior.

A: Yes, you are permitted to do so under the AGPL-3.0 License. However, the "Network Interaction" clause of the AGPL states that if you run a modified version of this script on a server (SaaS), you must make your modified source code available to the users of that service.

A: No. The KNN classification is calculated on bar closes. Once a bar is confirmed and the signal is printed, the historical pattern matching for that specific point in time remains fixed. This helps align backtesting results with bar-close behavior.

This project integrates concepts from behavioral finance, machine learning, and quantitative risk management. Below is the list of foundational research papers and literature used to design the QIA (Quant Investment Assistant) logic.

• Statman, M. (1995). "A Behavioral Framework for Dollar-Cost Averaging." The Journal of Portfolio Management, 22(1), 70-78.
  • Explores why investors prefer DCA for psychological reasons (regret minimization) despite mathematical sub-optimality in bull markets.
• Thorley, S. R. (1994). "The Fallacy of Dollar-Cost Averaging." Financial Practice and Education, 4(2), 17-26.
  • Analyzes the mathematical properties of DCA compared to Lump Sum investing.
• Leggio, K. B., & Lien, D. (2001). "Does Dollar Cost Averaging Make Sense?" Financial Services Review, 10(1), 73-86.
  • A critical assessment of DCA performance using risk-adjusted return metrics (Sortino/Sharpe).
• Cui, R., & Zhang, Y. (2023). "SmartDCA: An Enhanced Dollar-Cost Averaging Strategy with Deep Reinforcement Learning." arXiv:2308.05200.
  • Proposes dynamic position sizing adjustments to traditional DCA using reinforcement learning. Inspired DCAi's adaptive position weighting and Savings Pot concepts.

• Fix, E., & Hodges, J. L. (1951). "Discriminatory Analysis: Nonparametric Discrimination: Consistency Properties." USAF School of Aviation Medicine, Randolph Field, Texas.
  • The seminal paper that introduced the K-Nearest Neighbors (KNN) algorithm.
• Jaqobs (TradingView). (2022). "Machine Learning: Lorentzian Classification." TradingView Public Library.
  • Pioneering open-source implementation of KNN with Lorentzian distance metric for financial pattern recognition. Directly inspired DCAi's ML engine architecture and feature engineering approach.
• De Prado, M. L. (2018). Advances in Financial Machine Learning. Wiley.
  • The industry standard reference for applying ML techniques to financial time-series data.
• Jansen, S. (2020). Machine Learning for Algorithmic Trading. Packt Publishing.
  • Practical implementation of ML strategies in trading systems.

• Wilder, R. S. (1978). New Concepts in Technical Trading Systems. Trend Research.
  • The origin of the Average True Range (ATR), which represents the volatility component (Feature f3) in the QIA machine learning model.
• Quong, G., & Soudack, A. (1989). "Volume-Weighted RSI: Money Flow Index." Technical Analysis of Stocks & Commodities, 7(3).
  • The original publication introducing the Money Flow Index (MFI), used here as the primary momentum/volume oscillator for identifying capitulation.
• Murphy, J. J. (1999). Technical Analysis of the Financial Markets. Penguin York Institute of Finance.
  • The standard textbook for technical analysis, providing the definitive definition of Rate of Change (ROC) as the purest measure of price velocity and momentum used in Feature f2.
• Harris, L. (2003). Trading and Exchanges: Market Microstructure for Practitioners. Oxford University Press.
  • Provides the theoretical framework for Cumulative Volume Delta (CVD) by analyzing the bid-ask spread and aggressive order flow (Delta) to identify buyer/seller absorption.
• Hosoda, G. (Ichimoku Sanjin). (1969). Ichimoku Kinko Hyo (一目均衡表). Economic Statistics Research Institute.
  • The original source for the Ichimoku Cloud theory, used in this system to define the equilibrium zones for trend validation.

• Sortino, F. A., & Price, L. N. (1994). "Performance Measurement in a Downside Risk Framework." The Journal of Investing, 3(3), 59-65.
  • Introduction of the Sortino Ratio, distinguishing between harmful volatility (downside) and general volatility.
• Young, T. W. (1991). "Calmar Ratio: A Smoother Tool." Futures Magazine, 20(10), 40.
  • Introduction of the Calmar Ratio (CAGR / Max Drawdown) for evaluating hedge fund performance.

• Pardo, R. (2008). The Evaluation and Optimization of Trading Strategies. Wiley Trading.
  • Methodologies for backtesting, walk-forward analysis, and avoiding overfitting.
• Narang, R. K. (2013). Inside the Black Box: A Simple Guide to Quantitative and High-Frequency Trading. Wiley.
  • Insights into the structure of professional quant systems and alpha generation.

THIS IS NOT FINANCIAL ADVICE. This software and documentation are provided for educational and research purposes only. Nothing herein constitutes investment advice, a recommendation, or an endorsement of any security or investment strategy. The authors are not licensed financial advisors, investment advisors, or registered representatives. Do not rely on this software for investment decisions.

You use this software entirely at your own risk. Trading and investing in ANY asset class (crypto, stocks, indices, commodities) involve substantial risk of loss, including the potential loss of your entire principal. No trading strategy, algorithm, or machine learning model—including DCAi—can guarantee profits or prevent losses. Market conditions are unpredictable and can change rapidly due to:

• Geopolitical events and news shocks
• Technical platform failures (TradingView, exchanges, brokers)
• Flash crashes, circuit breakers, and execution slippage
• Regulatory changes and sudden liquidity withdrawal
• Black Swan events and unforeseen systemic risks

Under the AGPL-3.0 License, this software is provided "AS IS" without any warranty whatsoever. The authors explicitly disclaim:

• Any express or implied warranty of merchantability
• Any warranty of fitness for a particular purpose
• Any warranty that the software will be error-free or uninterrupted
• Any warranty regarding accuracy, completeness, or usefulness of results
• Any warranty about the performance of the KNN algorithm or ML predictions

The authors shall not be liable for any direct, indirect, incidental, special, consequential, or punitive damages, including but not limited to:

• Financial losses or costs incurred from using or relying on this software
• Lost profits, loss of investment capital, or opportunity costs
• Data loss or corruption
• Third-party claims or regulatory fines
• Any other damages arising out of or in connection with this software

This limitation applies regardless of whether damages were foreseeable or whether the authors were advised of the possibility of such damages.

• Historical results do not guarantee future performance. Backtesting uses historical data and cannot account for future market conditions, regime changes, or structural breaks.
• Backtesting bias: Model parameters may be overfit to past data. KNN models can have reduced predictive power in novel market conditions.
• Execution reality: Simulated results assume perfect order execution. Real-world trading incurs slippage, fees, commissions, and liquidity constraints.
• No repainting disclaimer: While the KNN model does not "repaint," TradingView's bar aggregation or time zone settings may cause signals to appear at different times than expected.

• Pine Script Limitations: This is a Pine Script v6 indicator running on TradingView. It is dependent on:
  • TradingView's data feeds, which may contain gaps, errors, or delays
  • The stability and availability of the TradingView platform
  • Correct bar aggregation and time zone settings
• Broker/Exchange Risks: Execution of trades based on DCAi signals depends on your broker's systems, liquidity, and regulatory compliance. DCAi has no control over broker order execution.
• Connectivity & Latency: Internet outages, platform downtime, or network delays may prevent signal execution.

• KNN is non-parametric: The algorithm relies on historical similarity. In truly novel market regimes, performance may degrade significantly.
• Feature engineering limitations: The three features (MFI, ROC, ATR) are normalized metrics. Extreme events (market halts, regulatory interventions) may break this model.
• Lookback window constraint: The model is limited to ~2,800 bars of history due to Pine Script computational constraints.
• No guarantee of convergence: The "4-bar forward prediction window" is arbitrary and may not capture the true reversal zone in all markets.

DCAi's behavior is highly asset-dependent. The pre-configured parameters (MFI thresholds, Rho sensitivity) may work well for the tested asset classes (Crypto, Stocks, Indices, Commodities) but:

• May not generalize to other assets (forex, futures, micro-caps)
• Require different tuning for different market conditions (bull vs. bear markets)
• May experience unexpected behavior on assets with low liquidity or unusual volatility profiles

You are solely and entirely responsible for:

• Your own investment decisions and capital allocation
• Conducting thorough due diligence before using this software
• Understanding the risks and limitations outlined above
• Monitoring and validating signals before executing trades
• Complying with all applicable laws and regulations in your jurisdiction
• Consulting with a qualified, licensed financial advisor before deploying capital

The inclusion of academic references in this README does not imply endorsement by those authors or institutions. Academic research is the foundation for concepts (DCA, KNN, Lorentzian Distance) but does not validate the specific implementation or results of DCAi.

By using this software, you acknowledge that you have read, understood, and agree to assume all risks outlined above.

1. Copy the Pine Script code into the TradingView Pine Editor.
2. Select your Asset Class in the settings to auto-configure volatility parameters.
3. Select the Start Date of your DCA
4. Monitor the Live Dashboard for real-time ML confidence and budget status.

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

• Commercial Use: You can use this for commercial purposes.
• Modification: You can modify the code.
• Source Disclosure: If you use this code in a web application or service (SaaS), you must release your source code under the same license.
• Attribution: You must keep the original copyright and license notice in all copies.