Either the gold bull market is over or it's not. If it's not, then gold will hit $2000. But when? I decided to use actual numerical data (monthly prices) instead of drawing lines on a goddamn chart.
Can we estimate a parameter that captures the price action for the entirety of the bull market thus far? That's 12+ years of data points, so assuming the bull market isn't over (and again, everything that follows here rests on that assumption), extrapolation becomes a decently safe bet.
Turns out the monthly closing price of gold since 2001 has increased at a remarkably steady rate. The data points somewhat surprisingly satisfy the main assumptions of a (log) linear regression model (which can be checked by various diagnostic measures that I won't get into). The one violation is that prices constitute a time series, so the *signs* of month-to-month differences will be correlated, but there are fairly dependable ways to correct for that when they're not too glaring.
How well does the model perform? The 99% confidence interval for the expected monthly change is remarkably tight: between +1.042% and +1.050%. The 99% confidence interval for each month's price is depicted as a capped error bar (grey). At the center of each of those error bars you can imagine the best point-estimate for price that month. The yellow band covers the 99% confidence interval for what that best estimate is, each month. It's fairly wide, mainly because I used a more conservative calculation of standard deviation that takes into account that we're dealing with a time-series, and prices are correlated.
The green dotted line depicts when gold is expected to close a month at $2000 (June 2013); the blue dotted line depicts the most conservative estimate for when price is expected to close a month at $2000 (November 2013); and the purple dotted line depicts when the lower bound of the 99% confidence interval passes $2000 (August 2014). Note that the upper bound of the month-to-month confidence intervals passed $2000 only in May 2012.
Another way of making linear regression more amenable to a time series is to take averages and use intervals. On the following chart, I use quarterly data, namely, the average closing price of the 3 months per quarter. That satisfies the linear regression even better, but the sample size is smaller, and the predictions are similar. Note that for the quarter starting this April and ending in July, it would be outside the predicted 99% confidence interval if gold's average monthly close lands below $1600.
I'd say that if price falls out of these intervals, then (and probably only then) should anyone say the bull market in gold may be over.