Uncovering Hidden Insights in Gold and Silver Market Charts with AI

 

AI can detect subtle patterns which the human brain may not be able to. Here are three examples. In these examples the AI converts the charts into pixels as well as uses OCR (Optical Character Recognition) to quantify changes in shades of colours, measure ‘distances’ between lines, rates of change and so on. Here are some of the hidden insights it uncovered. 

1. FAN CHARTS 

The fan charts above showing forecasted prices of gold and silver 60 days ahead were generated after 10,000 trials of a Monte Carlo Simulation. Q stands for quantile thus Q90 is 90th quantile. Gold's Q90–Q75 spread (47.6 points) is in fact wider than its Q25–Q10 spread (36.0 points).  Gold's dollar distribution is modestly top-heavy: about 21.7% of upside room to Q90 against 18.2% of downside room to Q10, measured from the median. Silver's is extreme: +59% of  room above the median versus −34% below. The model isn't merely uncertain about silver; it is saying that if it is wrong, it is far more likely to be wrong on the upside. 

PINBALL LOSS CHART

Pinball loss (also called quantile loss) is how you grade a forecast that comes as a range rather than a single number. Each quantile line is scored with a deliberately lopsided penalty: the Q90 line is penalised more haevily for underestimation, while the Q10 line is penalised more heavily for overestimation. Lower scores mean better-placed lines.

At 60 days, the Q90 loss is running at roughly two to two-and-a-half times the Q10 loss for both metals, and the Q90 bar sits almost level with Q50 when it should be comfortably below it. Translation: realized prices have repeatedly broken through the model's upper band, while the lower band has barely been tested. The bull market has consistently outrun even the model's 90th-percentile optimism. 

SKEW SCORE HISTORY CHART

The skew score answers the question how much room is there above the median versus below it? A perfectly symmetric forecast reads zero, one with its width above the median reads positive (bullish lean), and one with its width below reads negative (bearish lean). The weekly skew history is a running record of which way the model's imagination is leaning each week. 

Skew score is measured in log returns whereas the fan charts are in dollars . For Silver the skew score is 0.465 of upside against 0.417 of downside. For Gold: upside IS 0.197 up against ln downside of 0.201 down.  

The model's return-space lean is bearish gold, bullish silver — and it has been for weeks. Gold’s mildly bearish return-space lean argues for selling into strength without much fear of leaving money on the table, while silver's genuine positive skew compounding adjustment — says the cost of selling too early is higher.  In general, over the long term, the model has persistently under-forecast strength, so its bearish gold lean needs to take into account this factor too if you are a long terem gold investor.