Spot Silver Forecast for 27 Oct to 21 Nov 2025

 A big Thank You to EODHD.com for the data without which this post could not have been written.

The Close of Silver (XAGUSD) as of Friday 24 Oct 2025 is 48.62.The 21 business days ahead forecast for spot investment grade Silver bullion is summarized thus: 

10% chance of going below 48.32

50% chance of being 54.33

10% chance of being above 60.80

Methodology

We prefer to apply probabilistic quantile forecasting due to the highly unstable situation as the MAGA Man keeps changing his policies almost daily, and even just the words he uses generates  asymetric shocks to the model that renders straight forecasts useless. 

Model input variables: 

• 10yr Treasury yield
• US Dollar Index
• Spot Gold price
• S&P500
• US Unemployment Rate

Methodology

Methodology summary (plain-language bullets)

• Data we used

  • Daily end-of-day prices from EODHD for Silver (XAGUSD) and Gold (XAUUSD).
  • Macro/market context series (Dollar Index, 10-year Treasury yield, S&P 500, unemployment).
  • We reindexed everything to business days and forward-filled missing values to keep the timeline aligned.

• What we’re forecasting

  • The 21-business-day move in Silver, expressed as a return (log return), then translated back into a price level to make it intuitive.
  • We generate three scenarios at that horizon: a low case (q10), a median case (q50), and a high case (q90).

• Features the model looks at

  • Recent moves in Silver and Gold (daily returns).
  • Macro moves (Dollar Index, 10-year yield, S&P 500, unemployment — daily changes).
  • Silver’s own technical context: short- and medium-term moving averages (5- and 20-day), 20-day volatility, 20-day momentum, and the Gold/Silver ratio.

• Preprocessing and compression

  • We standardize features so no single scale dominates (everything measured on a comparable scale).
  • We use PCA to compress correlated information into a smaller set of components that retain most of the signal while reducing noise.

• Quantile forecasting (what and why)

  • Instead of predicting one number, quantile forecasting estimates price thresholds at chosen probabilities:
    • q10: a conservative “low-case” (only ~10% of outcomes are expected to be below this).
    • q50: the median (about half of outcomes above/below).
    • q90: an optimistic “high-case” (only ~10% of outcomes are expected to be above this).
  • Why use it? Markets are uncertain and asymmetric. Quantiles give a range with probabilities, helping you think in scenarios (downside risk, typical outcome, upside potential) instead of a single-point guess.

• Pinball (quantile) loss — how it works in simple words

  • For a given quantile, the model is penalized differently for over- and under-shooting:
    • At q90, under-predicting big up moves hurts more than over-predicting, nudging the model to place that line higher so only ~10% of cases exceed it.
    • At q10, over-predicting hurts more than under-predicting, nudging the model lower so only ~10% of cases end below it.
  • This asymmetric “pinball” penalty shapes each quantile line so the fraction of outcomes below (or above) matches the target probability.

• The modeling recipe

  • Train gradient boosting regressors separately for q10, q50, and q90 using pinball loss at each target quantile.
  • Feed the PCA-compressed features to the models.
  • At the forecast date, transform current features and get the three quantile returns, then convert them to prices by applying them to the current spot price.
  • Visualize as a shaded fan: a band from q10 to q90 with the q50 line in the middle, plus numeric labels at the anchor date for clarity.