Why the decline curve runs low: wells that bounce back

November 2026Illustrative, not for reserves or investment decisions

In the first backtest every group of wells was forecast a little low, and nearly all of that miss sat in wells whose rate went back up after the training window. Here we look closer: how big is the bounce, and can anything in the history before the forecast warn you? We reuse the same wells and fits as the first post.

1. How big is the bounce?

A well counts as recovered if its average rate over the 12 test months was higher than its average over the last three months of history.

Group	Wells tested	Share recovered	Median lift, recovered wells
Alberta oil	460	41%	+13%
Saskatchewan oil	469	35%	+15%
Saskatchewan gas	471	50%	+7%
BC gas	483	39%	+13%

Lift = average test-window rate vs the average of the last three months of history, median across recovered wells.

Between 35 to 50% of wells recovered, but the typical lift was modest: 7 to 15% above the last three months. A recovery does not need to be dramatic to leave a declining-curve forecast low, because the curve keeps falling while these wells flatten out.

2. A low month just before the forecast

One thing you can see at forecast time is a dip in the history. We flagged a well if its lowest month in the last six months of history was below half of its median rate over the last twelve. That is a rough sign of downtime, a curtailment or a data gap right where the curve is anchored.

Group	Wells with a low month	Median error: with a low month	Median error: without
Alberta oil	14%	−24%	−9%
Saskatchewan oil	13%	−26%	−7%
Saskatchewan gas	12%	−29%	−11%
BC gas	11%	−38%	−7%

Figure 1. Median forecast error on the 12-month volume, for wells with and without a low month in the last six months of history. Real data, same wells as the first post.

Only 11 to 14% of wells had such a month, yet for them the typical forecast ran 24% to 38% low, two to five times the miss of the other wells (7% to 11% low). A fit that starts from a depressed final stretch tends to extrapolate a rate that is too low.

3. The dip is only part of the story

Recovered wells also ran low when there was no low month in the history, so the dip does not explain everything.

Group	Recovered, with low month	Median error	Recovered, no low month	Median error
Alberta oil	30	−42%	157	−22%
Saskatchewan oil	35	−39%	127	−23%
Saskatchewan gas	34	−37%	203	−15%
BC gas	38	−50%	149	−20%

Figure 2. Distribution of forecast errors, all four groups pooled. Wells that kept declining cluster around zero; wells that recovered spread toward under-forecasts. Real data.

Wells that kept declining are centred on zero error. Recovered wells are shifted toward under-forecasts, with or without a visible dip. We cannot see why a well recovered from this data; restarts, workovers, new completions and operating changes would all look alike here.

What this means in practice

Before trusting a fit, look at the last six months of history. A low month right at the end is a warning that the forecast will probably run low.
Treat a recovery allowance as a normal part of a portfolio forecast: 35 to 50% of wells in these samples ended up above their recent rate.
We have not tested a fix. Two obvious experiments are fitting with the dip months removed, and anchoring on a median instead of the last months. They are candidates for a future post.

Limits

Same sample as the first post: up to 500 wells per group that kept producing, so wells that shut in for good are excluded.
The recovered flag uses the test window, which you would not have at forecast time. The low-month flag uses history only, so it is the one you could act on.
The half-of-median threshold is a rule of thumb we chose; we did not tune it, and other thresholds would give different counts.
One time window (January 2022 to September 2026).
The Alberta oil group includes wells licensed for crude bitumen; we did not separate them.

Reproduce it

Code: github.com/Jay-Kimani/TidyWells, posts/recovery.py. Download the files yourself from Petrinex and the BC Energy Regulator; we do not host the data.

Illustrative only. Not for reserves reporting or investment decisions.

Contains information licensed from Petrinex public data (Alberta, Saskatchewan) and the BC Energy Regulator; Crown copyright of the provinces is acknowledged. Tidywells is not affiliated with any regulator.