Gold/Silver Weekly Update 17/10 plus Quant Lesson

 

I do my weekly updates on gold/silver prices based on distilling the output of my 7 AI-driven models and deriving a conclusion based on my domain expertise. The models are:

1. GBDT-GARCH- My base model combining ensemble of gradient-boosted Decision Trees, GARCH, Monte Carlo and Quantile Pinball Loss. 

2. DXY Monitor- tracking USD against EUR and JPY

3. Real Yields Monitor- tracking Nominal Interest Rates minus Inflation

4. JPY Carry Trade Unwind Monitor-Treasuries/JGB yield gap, FX stress, intervention risk, position crowding, BOJ and Finance Ministry news.

5. Hedge funds leveraged basis trade risk monitor- what those devils in the Cayman Islands are doing

6. Private credit, corporate bonds, AI debt risk monitor: A problem because they comete for Treasuries for funds

7. Gold/Silver Headwinds Monitor: Probabilities on which of the 6 factors above will break first 

For this week, we analyse two charts from my models. The forecast probability chart and the skew distribution charts both from the GBDT model plus a table to explain and resolve the contradiction between the two outputs.

Reading the 4-panel skew chart

• Skew is the leaning to the left or the right and changes in the thickness length of the tails of a distribution. 
• Each panel shows the fitted skew-normal distribution approximated from  Q10/Q50/Q90 values. The solid line is Week 1 (10 Oct), the dashed line is Week 2 (16 Oct). Vertical lines mark Q10 (red), Q50 (white), Q90 (green).
• Gold 20-Day — Bullish lean, skew moved from +0.073 → +0.115. The distribution shifted slightly left (lower prices overall as spot fell from 4220 to 4170) but the upside tail got fatter — the model sees more asymmetric upside potential even as gold dipped. Prob Up: 64%.
• Gold 60-Day — Bearish lean, skew deepened from −0.123 → −0.139. The fat downside tail arrow confirms it. Despite the model’s 82.8% Prob Up, the distribution is pulled left — meaning the model thinks price is more likely up, but if it goes wrong, it goes badly wrong on the downside. This is the key tension in your report.
• Silver 20-Day — The most dramatic shift: skew jumped from +0.065 → +0.263. The dashed curve has noticeably shifted rightward — a much fatter upside tail. This is a strong bullish lean signal, even as silver spot fell from 61.11 to 56.84. The model sees the dip as buying opportunity asymmetry.
• Silver 60-Day — Flipped from −0.042 (bearish) to +0.051 (just crossing into bullish territory). A meaningful sign reversal in one week.

How predictive is skew for price direction?

• Moderately useful — but not as a standalone signal. Think of it as a lean indicator, not a trigger.
• Skew tells you about distributional asymmetry — which tail is fatter — not about the most likely outcome. That’s already captured by Q50 and Prob Up. The specific value of skew as a predictor:
• Over many studies of options-implied skew (the closest real-world analogue), directional accuracy is roughly 55–62% — better than a coin flip, but not dramatically so.
• The GBDT skew is a model output, not market-implied, so it reflects what the feature set is seeing about asymmetry in historical patterns. It will lag fast-moving events.
• The direction of skew change week-on-week is often more informative than the level. Silver 20D flipping from near-zero to +0.263 in one week is a stronger signal than a static +0.10.
• Skew is most reliable when it agrees with Prob Up and Q50 direction. When it disagrees — as with Gold 60D right now — that’s the interesting case.
• If skew contradicts the rest of the report, how much to discount?
• Gold 60D is a good example: Prob Up = 82.8%, Q50 points well above spot, yet skew = −0.139 (bearish). How to think about this:
• The contradiction is not a flaw — it is information. It means: “The model expects price to go up, but if it’s wrong, the downside scenario is more severe than the upside scenario is generous.” In plain terms: asymmetric risk to the downside, even in a bullish base case.
• Gold 60D currently sits at −0.139 with bullish Prob Up — I’d call that moderate contradiction, suggesting you hold the bullish view but with slightly smaller size or a tighter downside stop. You would not throw out the 82.8% Prob Up figure; you would respect that a bad outcome could be sharper than a good outcome.
• Silver 20D at +0.263 agrees with a modestly bullish Prob Up (56.7%) — that’s actually the cleanest signal this week. Low contradiction, fat upside tail, price recently dipped: the model is leaning hard into a bounce.

Actions table for skew and probability readings

Bottom line for this week: Silver 20D is the highest-conviction setup (skew and Prob Up aligned, strong bullish lean after a dip). Gold 60D is the most cautious setup (bullish base case but meaningful downside tail risk). Gold 20D and Silver 60D are in between, both with mild to moderate bullish lean.