Decline Curve Analysis Explained: How to Model Production

A Forecast Is the Only Thing You Are Really Buying

An oil well is a wasting asset with a known shape. It produces hardest on its first day and declines from there. Decline curve analysis (DCA) puts a number on that shape: fit a curve to what the well has already done, extend it forward, and read off two things. How much it produces next year, and how much it produces in total before it is plugged. That total is the estimated ultimate recovery, or EUR.

Nearly everything else in E&P analysis sits downstream of that curve. Reserves are the curve, certified by an auditor. Finding and development cost is capital divided by the volume the curve predicts. A DCF discounts the cash flows the curve generates. Get the curve wrong and every metric built on it is wrong in the same direction, which is what makes DCA worth an hour of anyone’s time.

The maths has been settled since Arps published it in 1945 and takes ten minutes to learn. The judgement is where the money is.

The Three Primary Decline Curves

Exponential Decline

Exponential decline is the simplest form. Production falls by a constant percentage each period:

Q(t) = Q₀ × e^(−D×t)

Where:

• Q(t) = production at time t
• Q₀ = initial production rate
• D = nominal decline rate, a decimal per year
• t = time in years

Characteristics:

• One parameter to fit besides the starting rate, so it is the easiest curve to work with
• Typical rates: 30–80% a year early in an unconventional well, 15–30% for a conventional one
• Plots as a straight line on semi-log paper, which is why it survived the era before computers
• Every year removes the same percentage, so the curve never flattens

Example: a well starting at 500 BOE/day and losing 30% of its rate each year:

• Year 1: 350 BOE/day
• Year 2: 245 BOE/day
• Year 3: 172 BOE/day

One wrinkle, because it trips people up in a spreadsheet. That 30% is the effective decline, the drop you would measure off a production chart a year apart. The D inside the exponential is the continuous equivalent, 0.357, because a rate falling smoothly all year ends lower than one stepped down once at the end. Quoted decline rates are almost always effective ones.

Exponential decline is the pessimistic case. It keeps taking the same percentage forever, so it drives the rate towards zero faster than real wells go. Most wells flatten into a long tail instead, and that is what the other two curves are for.

Hyperbolic Decline

Hyperbolic decline is the most widely used form, because it lets the decline rate change over time:

Q(t) = Q₀ × [1 + (D₀ × b × t)]^(−1/b)

Where:

• Q₀ = initial production rate
• D₀ = initial decline rate (at time t=0)
• b = hyperbolic exponent, a number between 0 and 1 for a well-behaved fit
• t = time in years

The b-factor sets the curve shape, by deciding how fast the decline rate itself decays. At b = 0 the decline rate never decays and the formula collapses back to exponential. At b = 1 you get harmonic decline, the flattest of the three. Values in between trade steepness now for tail later.

Characteristics:

• Captures the early steep drop and the flattening that follows, which is what real wells do
• Needs three parameters (Q₀, D₀, b), so it needs more history to pin down than exponential
• The extra parameter, b, is also the one nobody can see directly in the data for years

Example: a well starting at 800 BOE/day with D₀ = 0.50 and b = 0.4:

The right-hand column is the point. The rate keeps falling, but it falls more gently every year. That easing is the whole reason hyperbolic curves are used, and it is also where the trouble starts.

The b-Factor Is Where Reserves Get Invented

Integrate the hyperbolic curve to get cumulative production and it converges to a finite number only while b is below 1. At b = 1 exactly, and at anything above it, the integral diverges: the maths says the well produces forever and the EUR becomes whatever horizon you happened to stop calculating at. Push b to 1.3 and reserves rise without any geology changing.

This is not a textbook curiosity. Fit Arps to the first two or three years of a shale well and the best-fit b routinely comes back above 1, because ultra-low permeability rock stays in transient flow for years and the early data really is that flat. Take that fit at face value and you have booked an unbounded reserve.

The standard defence is a terminal decline switch: run the hyperbolic while it fits, then convert to exponential once the decline rate eases to a floor, commonly 5 to 10% a year. Reserve auditors do this as a matter of course. If someone hands you an EUR and cannot tell you the terminal decline assumption behind it, they have handed you a number with no ceiling.

Harmonic Decline (b = 1.0)

Harmonic decline is the boundary case, where the decline rate falls in direct proportion to the rate itself:

Q(t) = Q₀ / [1 + (D₀ × t)]

It produces the longest tail of the three, which suits conventional fields that produce at low rates for decades. It is also the point at which cumulative production stops converging, so a harmonic forecast has to be cut off at an economic limit rather than run to infinity.

Initial Production Rate (IP) and Well Quality

Every decline curve has to start somewhere, and that somewhere is the initial production rate. For a shale well it is normally quoted as IP30: the average daily rate over the first 30 days on production. It is the earliest hard read anyone gets on whether the rock and the completion were any good.

Rough shale bands, for orientation rather than valuation:

• Core acreage, long lateral: 1,000–1,800 BOE/day
• Average acreage: 600–1,000 BOE/day
• Edge of the play, or a short lateral: 300–600 BOE/day

Note the words “long lateral”. IP scales roughly with how much rock the wellbore touches, so a 15,000-foot well will out-produce a 7,500-foot well on the same acreage and tell you nothing about the geology. Operators normalise IP per 1,000 feet of lateral for exactly this reason, and comparisons that skip the normalisation flatter whoever drilled longer.

IP trends across a programme matter more than any single well. Average IPs falling 15% year on year points at degrading acreage, tighter well spacing draining the same rock twice, or a completion design that has stopped improving.

Estimated Ultimate Recovery (EUR)

EUR is the total volume a well is expected to produce over its economic life: the area under the decline curve, from first production to the point where the well stops paying to keep open.

EUR = ∫ Q(t) dt from t=0 to t=T_economic

That end point is an economics question, not a geology one. Lease operating costs run to a few thousand dollars a month on a typical onshore well, so at a $70/bbl planning price the cut-off lands at only a handful of barrels a day, and it rises as prices fall. A price crash retires marginal wells by moving the economic limit, not by changing the reservoir.

For a shale well the shape of the curve, and above all the b-factor, does most of the work in setting EUR, because the tail years are numerous even though each contributes little. Two engineers fitting the same three years of data with different b assumptions routinely land 20 or 30% apart on the total. See How to Read a Reserves Report for how an EUR becomes a booked reserve.

Type Curves by Basin

Basins decline differently because the rock and the completion practices differ. A “type curve” is the industry’s shorthand for how a representative well in a given area behaves, and it is what operators use to justify a drilling programme.

Read published type curves with your guard up. An operator draws its own, from the wells it chooses to include, on the acreage it wants you to value, and a type curve is a marketing document at least as much as an engineering one. The tell is usually what has been left out: wells excluded as non-representative, a lateral length quietly longer than the field average, a tail assumption doing more work than the data supports. The profiles below are illustrative shapes, not any company’s disclosure.

A quoted b depends heavily on how much history the fit used and on where the terminal switch sits, so the published bands below are not directly comparable with each other or with a fit you do yourself. Early-life fits come back high, sometimes above 1. Bands quoted off several years of data come back lower. Same wells, different number.

Permian Basin (Tight Oil)

Characteristics:

• IP30: 1,000–1,800 BOE/day, depending on acreage quality and lateral length
• Decline: 65–85% in year one, 30–50% across years two and three, easing towards 10–15% by year ten
• EUR: broadly 250,000–450,000 barrels of oil per 10,000-foot lateral. On a BOE basis the figure is much larger, because Permian wells make a great deal of associated gas and NGLs alongside the oil

That last bullet is the one people get wrong. Permian EURs are quoted both ways, and an oil EUR set against a BOE production profile understates the well by something like half. Establish which one you are looking at before you compare it with anything.

Type Curve Dynamics:

• Longer laterals and denser completions have lifted headline IPs over the past decade, but much of that gain is more rock per well rather than better rock
• Tighter well spacing cuts the other way: wells drilled close together drain each other, and “parent-child” interference shows up as weaker results on the later wells

An illustrative Permian profile, a hyperbolic fit at b = 0.9 starting from 1,200 BOE/day:

• Year 1: 350 BOE/day (71% down)
• Year 2: 200 BOE/day (43% down)
• Year 3: 140 BOE/day (30% down)
• Year 5: 80 BOE/day
• Year 10: 40 BOE/day

Note the b of 0.9, close enough to the harmonic boundary that the tail has to be handled deliberately rather than extrapolated. Switch to an 8% terminal exponential around year thirteen, where the rate is near 30 BOE/day, and the curve produces roughly 700,000 BOE over a producing life of about thirty-five years. Leave it unswitched and the same curve settles at about 1.9 million BOE, nearly three times as much, on a tail that would still be dribbling out barrels in the twenty-second century. One assumption, made off-screen, moves the answer more than the drilling programme does.

Bakken Shale (North Dakota)

Characteristics:

• IP30: 1,000–1,600 BOE/day, highly variable by landing zone and completion
• Decline rate: 70–90% year one, then 35–50% across years two and three
• b-factor: 0.35–0.55 on fits with several years of history
• EUR: 200,000–350,000 barrels of oil per 10,000-foot lateral

Bakken versus Permian: the Bakken drops harder in year one and recovers less oil per well. It is also a single stacked play rather than the Permian’s several benches, so an operator gets fewer cracks at the same acreage. Lower permeability and heavier fluid drive the steeper profile.

Eagle Ford Shale (Texas)

Characteristics:

• IP30: 900–1,500 BOE/day on oil-leg wells; gas-leg wells lower
• Decline: 60–80% year one, then 25–40% across years two and three
• b-factor: 0.4–0.6 on fits with several years of history
• EUR: 180,000–320,000 barrels of oil per lateral on the oil leg; 3–6 billion cubic feet of gas per well on the gas leg

Zoning variability: the Eagle Ford spans several landing zones, Upper, Lower, and the Austin Chalk below, each with its own pressure, fluid properties and well performance. A type curve that does not name its zone describes nothing in particular.

Conventional Onshore Fields

Characteristics:

• Lower IP: 100–400 BOE/day per well, depending on reservoir pressure and permeability
• Much gentler decline: 5–15% a year, exponential or harmonic
• b-factor: 0.7–1.0, close to the harmonic end
• EUR: 500,000–3,000,000 BOE per well, far more than an unconventional well recovers, over a much longer life
• Economic tail: 20 to 40 years at low rates

Example: Conventional Gulf of Mexico field well

• Year 0: 250 BOE/day
• Year 1–5: 8% annual decline
• Year 6–20: 3–5% annual decline
• Year 21+: 1–2% annual decline, until abandonment costs outweigh what the well still earns

Deepwater Offshore

Characteristics:

• Pressure and rock quality give a high IP: 1,000–5,000+ BOE/day per well
• Moderate initial decline: 15–30% in year one
• Gentle long-term decline of 3–10% a year, much like a conventional field
• EUR: 10–100+ million BOE per field
• Long producing life: 20–40 years

A big, high-pressure reservoir holds its rate up as it depletes, which is why deepwater declines gently. Unconventional rock is tighter, so flow falls away with the pressure.

Fitting a Curve in Practice

Fitting Data

To fit a decline curve, you need:

1. Historical production data: monthly, ideally 24 to 36 months, and never fewer than 18 for a hyperbolic fit
2. Initial conditions: spud date, completion date, and the date production started
3. Operational events: shutdowns, workovers or rate changes, which the data has to be adjusted for

Then fit the curve, one of three ways:

• Graphical method: plot rate against time on a semi-log scale. Exponential decline falls as a straight line, so any curvature tells you the decline is hyperbolic and roughly how much
• Regression: least squares on Q₀, D₀ and b together. Be aware that several combinations fit early data almost equally well and imply wildly different EURs, which is the b problem showing up as a numerical one
• Type curve matching: compare against published curves for the basin, adjusted for acreage quality and completion design

DCA Pitfalls

1. Too little data: six months of history will not pin down a hyperbolic fit. Treat 18–24 months as the floor.

2. Extrapolation beyond data: a curve fitted to eighteen months of a shale well is being asked to predict thirty years, using data drawn from the steepest and least representative stretch of its life. Nearly all the volume in the forecast sits outside the period any data covers, and the parameter controlling that volume, b, is the one the short history constrains worst.

3. Ignoring rate constraints: surface infrastructure, gas lift and compression often hold early production below what the well could deliver. Fit that data raw and the curve describes the facilities rather than the reservoir.

4. Mixing wells: average two wells with different completion designs, landing zones or rock and you get a curve that describes neither.

5. Not stress-testing EUR: a single EUR implies a precision the method does not have. Run low, base and high cases, and vary the b-factor between them rather than only the starting rate.

EUR Is Not the Same as Reserves

An EUR is an engineer’s estimate of everything a well will eventually produce. Proved reserves are the subset of that an auditor will certify to a reasonable-certainty standard, priced by a rule rather than by judgement. Under the SEC regime the rule is the unweighted average of the first-day-of-month price over the trailing twelve months, which is neither spot nor the forward strip. So one well carries a single EUR and several different reserve numbers, depending on the regime, the year, and how much of the tail the auditor is willing to certify.

The value follows the volume, but not one for one. A well recovering 250,000 barrels of oil at our $70/bbl planning price generates about $17.5 million of gross revenue across its life. Royalties take a share off the top, operating costs run for thirty years, the drilling capital went out at the front, tax takes a cut of what survives, and the remainder arrives so slowly that discounting removes a large part of it again.

The figure above is barrels of oil, not barrels of oil equivalent. Gas converts into BOE at six thousand cubic feet to the barrel, which is an energy equivalence and nothing like a value one: at a $3.00/MMBtu planning price those six Mcf are worth about $18 against the barrel’s $70. Running a gas-weighted EUR at an oil price overstates the revenue nearly fourfold.

Aggregate the same logic across a drilling programme and you get the reserve replacement ratio, new proved reserves booked in a year divided by the year’s production. Above 100% and the company is replacing what it produces; below it and the reserve base is shrinking. Because bookings come out of EURs, an operator that runs optimistic type curves can print a healthy RRR for several years before the wells settle the argument. Replacement is only as good as the cost of achieving it, which is what F&D benchmarks measure.

Where a Forecasting Error Actually Hurts

In a DCF of an E&P company, the production line comes straight out of DCA:

Annual Cash Flow = (Realised price after royalty − Cost per BOE) × Production − CapEx

What matters is not only how wrong the forecast is but where the error sits. Overstate the first two years and the damage lands in cash flows that are barely discounted, so the valuation moves close to one for one. Overstate the b-factor and the extra barrels appear twenty years out, where a 10% discount rate has already reduced a dollar to about fifteen cents; the EUR can be badly wrong while the NPV barely notices.

Which is why the b-factor does more harm to reserve bookings, F&D costs and acquisition prices than to a discounted valuation. It is a volume error, and volume metrics have no discount rate to protect them. See NAV vs EV/DACF for how a production forecast turns into a valuation.

One Well Is Not a Development Plan

In the Permian and the Bakken, acreage gets developed in phases over 5–10+ years. A decline curve describes a single well. Planning a programme needs three more things:

• Rig scheduling: how many wells and pads a year?
• Pad size and well spacing: three, four or five wells per pad, and how far apart?
• Type curve assumptions: does each well follow the same decline, or do later wells behave differently?

A typical Permian development plan might run:

• Years 1–3: 40 wells/year (ramping production)
• Years 4–8: 50 wells/year (plateau)
• Years 9–12: 30 wells/year (winding down)

Each vintage follows its own curve. Companies usually assume later vintages come in stronger on completion improvements; the counterweight is that the best acreage gets drilled first and later wells sit closer to their neighbours. Which force wins is an empirical question, and the answer shows up in the IP trend rather than in the plan. Portfolio-level DCA is what ties reserve life to capital allocation at operators like EOG Resources and ExxonMobil.

The Limits of Curve Fitting

DCA tells you what a well has done. It cannot tell you what happens when spacing tightens, when the operator changes its completion design, or when prices fall far enough to move the economic limit. The curve is a description of history wearing the costume of a forecast, and the most common mistake is fitting a clean line through noisy data and then treating the EUR that falls out as a fact.

Two habits protect you. Present low, base and high cases rather than a single number, because EUR uncertainty of 20 to 30% is normal even on a well-behaved fit. And whenever you are handed an EUR, ask for the b-factor and the terminal decline behind it. If nobody can produce them, the number was never bounded in the first place. The Oil & Gas Sector Primer covers how DCA-derived reserves feed into NAV and EV/DACF valuation.