SILVER XAG/USD 2020-2025 - Daily AI Feature Feed

What You Get

• ~1,510 trading days from January 2, 2020 to December 31, 2025
• 90 columns (3 metadata + 80 features + 7 forward labels)
• Single CSV file, chronological order
• One row per trading day — clean, ready for ML ingestion

Feature Groups (90 columns)

Why Silver Sentiment Is Different

• Safe-haven vs industrial tug-of-war — recession fears are simultaneously bullish (safe haven) and bearish (industrial slowdown); our AI captures which narrative is winning
• Gold-silver ratio — the gold/silver ratio is a key macro signal; when it widens, silver is undervalued relative to gold, and news coverage shifts accordingly
• Solar and EV demand — silver is critical for photovoltaic cells and EV electronics; green energy policy sentiment directly impacts silver's industrial premium
• COMEX positioning — silver futures are heavily traded by speculators; sentiment spikes often precede large positioning shifts
• USD inverse correlation — like gold, silver moves inversely to the dollar; our AI tracks Fed policy narrative and USD sentiment daily

Collection Methodology

1. News Collection: GDELT (Global Database of Events, Language, and Tone) provides 100+ silver-related articles daily from global financial media
2. AI Sentiment Analysis: Google Gemini processes each article batch, scoring overall macro sentiment and silver-specific sentiment on a -1.0 to +1.0 scale
3. Price Data: Yahoo Finance provides XAG/USD closing prices, cross-asset prices (gold, oil, copper, bonds, DXY, Bitcoin, SPY), and options data
4. Feature Engineering: 87 features are computed from raw inputs using numpy — technical indicators, sentiment derivatives, cross-asset correlations, macro signals, and forward labels
5. Quality Control: Automated audit checks coverage, NaN rates, column integrity, and data consistency

Complete Column Reference

# | Column | Type | Group | Description
---|--------|------|-------|------------
1 | date | date | Metadata | Trading date (YYYY-MM-DD)
2 | year | int | Metadata | Calendar year
3 | day_of_week | string | Metadata | Day name (Monday–Friday)
4 | price_close | float | Price Action | Silver closing price (USD/oz)
5 | price_return_1d | float | Price Action | 1-day return (%)
6 | price_return_3d | float | Price Action | 3-day return (%)
7 | price_return_5d | float | Price Action | 5-day return (%)
8 | price_return_10d | float | Price Action | 10-day return (%)
9 | price_return_20d | float | Price Action | 20-day return (%)
10 | price_log_return_1d | float | Price Action | 1-day log return
11 | price_dist_high_10d | float | Price Action | Distance from 10-day high (%)
12 | price_dist_high_20d | float | Price Action | Distance from 20-day high (%)
13 | price_dist_low_10d | float | Price Action | Distance from 10-day low (%)
14 | price_dist_low_20d | float | Price Action | Distance from 20-day low (%)
15 | tech_rsi_14 | float | Technical | RSI 14-day (0–100, >70 overbought, <30 oversold)
16 | tech_sma_5 | float | Technical | 5-day Simple Moving Average (USD/oz)
17 | tech_sma_20 | float | Technical | 20-day Simple Moving Average (USD/oz)
18 | tech_sma_5_dist | float | Technical | Distance from 5-day SMA (%)
19 | tech_sma_20_dist | float | Technical | Distance from 20-day SMA (%)
20 | tech_ema_20 | float | Technical | 20-day Exponential Moving Average (USD/oz)
21 | tech_ema_20_dist | float | Technical | Distance from 20-day EMA (%)
22 | tech_bollinger_pos | float | Technical | Position within Bollinger Bands (0=lower, 0.5=middle, 1=upper)
23 | tech_bollinger_bw | float | Technical | Bollinger Band width (%)
24 | tech_bollinger_squeeze | float | Technical | Squeeze indicator (1 = bandwidth in bottom 10th percentile, upcoming breakout)
25 | tech_macd | float | Technical | MACD line
26 | tech_macd_signal | float | Technical | MACD signal line
27 | tech_macd_hist | float | Technical | MACD histogram (positive = bullish momentum)
28 | tech_adx | float | Technical | ADX trend strength (>25 trending, <20 ranging)
29 | tech_roc_5 | float | Technical | 5-day Rate of Change (%)
30 | tech_roc_10 | float | Technical | 10-day Rate of Change (%)
31 | tech_roc_20 | float | Technical | 20-day Rate of Change (%)
32 | tech_streak | int | Technical | Consecutive up/down days (positive = up streak)
33 | tech_vol_5d | float | Technical | 5-day realized volatility (annualized %)
34 | tech_vol_10d | float | Technical | 10-day realized volatility (%)
35 | tech_vol_20d | float | Technical | 20-day realized volatility (%)
36 | tech_momentum_quality | float | Technical | Momentum consistency score (ROC adjusted for volatility)
37 | sent_overall | float | AI Sentiment | Overall macro sentiment (-1.0 bearish to +1.0 bullish)
38 | sent_silver | float | AI Sentiment | Silver-specific sentiment (-1.0 bearish to +1.0 bullish)
39 | sent_spread | float | AI Sentiment | Silver minus overall sentiment (positive = silver more bullish than market)
40 | sent_overall_ma3 | float | AI Sentiment | 3-day MA of overall sentiment
41 | sent_silver_ma3 | float | AI Sentiment | 3-day MA of silver sentiment
42 | sent_overall_ma5 | float | AI Sentiment | 5-day MA of overall sentiment
43 | sent_silver_ma5 | float | AI Sentiment | 5-day MA of silver sentiment
44 | sent_overall_mom3 | float | AI Sentiment | 3-day overall sentiment momentum
45 | sent_silver_mom3 | float | AI Sentiment | 3-day silver sentiment momentum
46 | sent_overall_mom5 | float | AI Sentiment | 5-day overall sentiment momentum
47 | sent_silver_mom5 | float | AI Sentiment | 5-day silver sentiment momentum
48 | sent_overall_vol5 | float | AI Sentiment | 5-day overall sentiment volatility
49 | sent_overall_vol10 | float | AI Sentiment | 10-day overall sentiment volatility
50 | sent_silver_vol5 | float | AI Sentiment | 5-day silver sentiment volatility
51 | sent_price_corr_10d | float | Sent x Price | 10-day rolling sentiment-price correlation
52 | sent_price_corr_20d | float | Sent x Price | 20-day rolling sentiment-price correlation
53 | sent_price_diverge | float | Sent x Price | Divergence flag (1 when sentiment and price disagree 3+ consecutive days)
54 | news_count | int | News Volume | Articles collected that day
55 | news_count_ma20 | float | News Volume | 20-day article count moving average
56 | news_count_zscore | float | News Volume | Z-score vs 20-day window (>2 = unusual volume)
57 | news_count_mom5 | float | News Volume | 5-day article count momentum
58 | news_spike | float | News Volume | Binary flag for abnormal news volume (count > 2x MA)
59 | mkt_equity_breadth | float | Cross-Asset | Equity market breadth proxy via SPY (0–100)
60 | mkt_equity_return | float | Cross-Asset | S&P 500 (SPY) daily return (%)
61 | mkt_commodity_dispersion | float | Cross-Asset | Cross-commodity return dispersion (gold, oil, copper)
62 | mkt_gold_return | float | Cross-Asset | Gold (GC=F) daily return (%)
63 | mkt_oil_return | float | Cross-Asset | Crude Oil (CL=F) daily return (%)
64 | mkt_copper_return | float | Cross-Asset | Copper (HG=F) daily return (%)
65 | mkt_bonds_return | float | Cross-Asset | Treasury Bonds (TLT) daily return (%)
66 | mkt_dxy_return | float | Cross-Asset | US Dollar Index (DXY) daily return (%)
67 | mkt_bitcoin_return | float | Cross-Asset | Bitcoin (BTC-USD) daily return (%)
68 | mkt_silver_vs_bonds | float | Cross-Asset | Silver return minus bond return (%)
69 | mkt_silver_vs_equity | float | Cross-Asset | Silver return minus SPY return (%)
70 | mkt_silver_equity_beta_20d | float | Cross-Asset | 20-day rolling beta of silver vs SPY
71 | vol_of_vol_20d | float | Vol Regime | Volatility of volatility (20-day rolling)
72 | vol_regime | int | Vol Regime | Regime: 0=low, 1=normal, 2=high, 3=extreme
73 | sent_vol_regime_interaction | float | Interaction | Silver sentiment x volatility regime (amplified signal in high-vol periods)
74 | macro_vix | float | Macro | VIX level (CBOE Volatility Index)
75 | macro_vix_change_1d | float | Macro | 1-day VIX percentage change
76 | macro_vix_ma5 | float | Macro | VIX 5-day moving average
77 | macro_yield_spread | float | Macro | 10Y minus short-term Treasury yield spread (%)
78 | macro_credit_spread | float | Macro | High-yield credit spread proxy (LQD/HYG ratio)
79 | macro_dxy_change | float | Macro | US Dollar Index 1-day percentage change
80 | options_iv_atm | float | Options | ATM implied volatility (%) via SLV beta approximation
81 | options_iv_rv_spread | float | Options | IV minus realized vol (positive = fear premium)
82 | macro_days_to_fomc | float | FOMC Calendar | Days until next Federal Reserve decision (0 = FOMC day, capped at 90)
83 | macro_is_fomc_day | float | FOMC Calendar | Binary flag: 1 on FOMC announcement day, 0 otherwise
84 | label_return_1d | float | Forward Labels | Next-day return (%)
85 | label_dir_1d | string | Forward Labels | Next-day direction (UP/DOWN/FLAT)
86 | label_return_3d | float | Forward Labels | 3-day forward return (%)
87 | label_dir_3d | string | Forward Labels | 3-day forward direction
88 | label_return_5d | float | Forward Labels | 5-day forward return (%)
89 | label_dir_5d | string | Forward Labels | 5-day forward direction
90 | label_flat_1d | float | Forward Labels | Flat flag (1 if next-day |return| < 0.3%)

Data Quality

• NaN values limited to first ~30 rows (indicator warmup period for rolling windows)
• Last 1–5 rows may have empty forward labels (not yet realized)
• Zero gaps in sentiment coverage — every trading day has article data
• Cross-asset features (gold, oil, copper, bonds, DXY, Bitcoin) sourced from Yahoo Finance; occasional NaN on mismatched trading holidays
• Quality rating: 5/5 (automated audit verified)

Known Limitations

• Silver-specific sentiment (sent_silver) is derived from AI analysis of English-language financial news via GDELT. Non-English sources and social media are not included.
• Options IV (options_iv_atm) is approximated using VIX x SLV beta, not from actual silver options chains. This is a standard quant approximation but may diverge from true ATM IV during extreme moves.
• Cross-asset returns require both assets to trade on the same day. Holidays in one market produce NaN for that pair.
• FOMC dates from 2026 onward are projected based on historical scheduling patterns and may shift slightly.
• The flat zone threshold in forward labels is 0.3% for silver (same as gold, reflecting commodity-level daily volatility).
• Silver's "silver screen", "silver lining", and similar non-financial references are filtered out, but occasional noise may remain in article counts.

Use Cases

• ML model training for silver price direction and magnitude prediction
• Sentiment-return relationship research across 6 years of macro regimes (COVID crash, inflation spike, rate hike cycle, dovish pivot, industrial recovery)
• Cross-asset correlation analysis: silver vs gold, silver vs dollar, silver vs copper (industrial proxy), silver vs crypto
• Gold-silver ratio modeling using dual sentiment and cross-asset features
• Volatility regime modeling for silver-specific dynamics (silver is historically more volatile than gold)
• FOMC event study: how sentiment shifts before and after Fed decisions
• Safe-haven vs industrial demand regime detection using sentiment divergence
• Feature engineering baseline for quantitative commodity strategies

Pairs Well With

License

AI Training Rights