FX Exponential Smoothing Dashboard

Streamlit workspace for researching and stress-testing a dual-exponential-smoothing FX trading strategy. The app ingests Yahoo Finance prices, applies configurable α/β smoothing, produces trade signals, backtests with transaction costs, and offers parameter optimization plus ML-based signal experiments.

Collaborators

• Tanish Patel tp2899@columbia.edu
• Aditya Pendyala ap4839@columbia.edu
• Rishit Belani rb3881@columbia.edu
• Arnav Ahuja aa5790@columbia.edu
• Aryamaan Srivastava as7845@columbia.edu

---

Table of Contents

• Concept in Brief
• Key Features
• Screens / Pages
• Project Structure
• Setup
• Running the App
• Strategy Logic
• Backtest & Metrics
• Presets
• Data & Symbols
• Troubleshooting
• Roadmap Ideas

---

Concept in Brief

• Two exponential smoothers: ES(α) (slow, trend) and ES(β) (fast, momentum).
• Signal driver: spread = ES(β) – ES(α). Crosses and thresholds generate longs/shorts; optional deceleration exit flattens on momentum loss.
• Simple enough for intuition, flexible enough to adapt across assets/timeframes.

Key Features

• Dual-smoothing with tunable α (slow) and β (fast), threshold x, and θ deceleration exit logic.
• Rich visuals: KPI cards, equity and drawdown curves, price with trade markers, return distributions, ES overlays.
• Trade log with filters and CSV export; turnover-aware transaction costs.
• Preset manager to save/load parameter sets across sessions.
• Grid-search page for α/β/x/θ surfaces with best-config selection.
• ML experiments page that fuses ES signals with a Random Forest classifier, feature importance, and optional SHAP interpretability.
• Works with major FX pairs by default; accepts any Yahoo Finance symbol (FX, crypto, metals, indices).
• Quick live tick pull to sanity check the latest price.

Screens / Pages

• fx-cross.py — Main Dashboard: parameter inputs, live pull, backtest, KPIs, charts, trade log, presets.
• pages/01_About_Strategy_Notes.py — Strategy Overview: quick description and contact.
• pages/02_Alpha_Beta_Optimization.py — Grid Search: sweep α/β with x/θ combinations, view heatmaps, and push best settings.
• pages/03_ML_FX_Signals.py — ML Hybrid Signals: Random Forest on engineered features + ES signals, performance diagnostics, SHAP (if installed).

Project Structure

• fx-cross.py — main Streamlit app (dashboard/backtest/presets).
• pages/01_About_Strategy_Notes.py — overview page.
• pages/02_Alpha_Beta_Optimization.py — grid-search UI and plots.
• pages/03_ML_FX_Signals.py — ML hybrid signals and explainability.
• utils/presets.py / utils/presets.json — preset I/O helpers and stored presets.
• requirements.txt — Python dependencies.

Setup

1. Create and activate a virtual environment (recommended):

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\\Scripts\\activate

2. Install dependencies:

pip install -r requirements.txt

3. (Optional) Install SHAP if you want explainability plots on the ML page and it fails to import automatically:

pip install shap

Running the App

Start the main dashboard:

streamlit run fx-cross.py

Streamlit will open in your browser. Use the left sidebar to switch pages or adjust parameters.

Strategy Logic

• Smoothing:
  • ES(α) = slow smoother; ES(β) = fast smoother.
  • Parameters α, β ∈ (0,1); β > α typically.
• Signal generation (generate_signals):
  • Long entry when spread crosses above +x from negative.
  • Short entry when spread crosses below –x from positive.
  • Optional deceleration exit: if Δspread < –θ, flatten position.
  • Signals are converted to positions via forward-fill of non-zero signals.
• Backtest (backtest):
  • Position-timed returns: positions.shift(1) * returns.
  • Transaction costs applied to turnover: positions.diff().abs() * cost.
  • Equity curve: cumulative product of (1 + strategy returns) scaled by starting capital.

Backtest & Metrics

Computed in backtest and surfaced in KPIs:

• Final Equity and Net Profit
• Total Return % and Annualized Return
• Volatility (annualized) and Sharpe Ratio
• Max Drawdown (absolute)
• Hit Rate (daily direction agreement with position sign)
• Number of Trades (unique entry signals)
• Trade log with Entry/Exit time, price, direction, PnL, Return %

Presets

• Stored locally in utils/presets.json; saved/loaded via the sidebar preset manager.
• A preset captures: symbol, timeframe, α, β, x, exit_rule, θ, cost, start/end dates.
• Safe to version-control but be mindful of personal preferences or experimental settings.

Data & Symbols

• Data source: Yahoo Finance via yfinance.
• Supported timeframes:
  • Main app: 1H, 4H, Daily, Weekly.
  • ML page: adds 1m, 15m, Monthly, Quarterly (subject to Yahoo data availability).
• FX examples: EURJPY=X, USDJPY=X, EURUSD=X, GBPUSD=X, AUDUSD=X, USDCAD=X, USDCHF=X, NZDUSD=X.
• Other assets: metals (XAUUSD=X), crypto (BTC-USD), indices/ETFs (any valid Yahoo symbol).
• Network access is required to fetch historical and live prices.

Troubleshooting

• No data returned / empty charts:
  • Check symbol spelling and timeframe support on Yahoo Finance.
  • Ensure start/end dates span at least a few bars.
  • Confirm network access.
• SHAP errors on ML page: install shap (see Setup step 3) or toggle off SHAP features.
• Beta slider disabled at low values: β lower bound is set to α + 0.01 to keep the “fast” smoother faster than the “slow” one.
• Preset not saving/loading: verify write access to utils/presets.json.

Roadmap Ideas

• Add walk-forward or rolling-window evaluation.
• Include position sizing and risk overlays (e.g., volatility scaling).
• Add portfolio view to combine multiple symbols.
• Expand ML page with alternative models and richer feature sets.
• Expose API endpoints for programmatic backtests.