Monte Carlo DCF Simulation

A Monte Carlo DCF simulation replaces fixed assumptions with probability distributions for revenue, margins, growth, and discount rates, then runs thousands of iterations to produce a range of possible valuations rather than a single point estimate. This approach reveals the distribution of outcomes and quantifies valuation uncertainty.

How Monte Carlo overcomes false precision

A traditional discounted-cash-flow-valuation model plugs fixed numbers into a formula and returns a tidy number—say, $150 per share. In reality, nearly every assumption carries uncertainty. Revenue could grow at 5% or 12%. Operating margins might compress from regulation. The discount-rate shifts with interest rates. A point estimate masks this scatter.

Monte Carlo simulation acknowledges that unknowns are unknowns. Instead of assuming “revenue grows at 8%,” you define a distribution: perhaps 60% probability of 6–10%, 30% of 10–14%, 10% of under 6%. You do the same for margins, capex ratios, terminal growth, and the discount rate. The algorithm then randomly samples from each distribution thousands of times, recalculating the valuation each iteration. The result is not “$150” but a probability distribution: “40% chance valuation exceeds $160; 50% chance it lands between $130 and $170.”

This is honest. It shows what you know and what you don’t.

The mechanics of random sampling

The simulation engine works in a loop:

1. Define distributions. For each input (revenue growth, EBITDA margin, capex-to-revenue, terminal-growth, discount-rate), specify the shape: normal (bell curve), triangular, uniform, or empirical (historical percentiles). Include correlations if some variables move together—higher growth often pairs with margin pressure.

2. Sample once. Draw one random value from each distribution. Not cherry-picked; genuinely random, respecting the shape you specified.

3. Run one full DCF. Feed these sampled values through your cash-flow-statement forecast, terminal-value calculation, and net-present-value computation. Write down the result.

4. Repeat. Go back to step 2. Do this 5,000, 10,000, or 50,000 times.

5. Analyze the distribution. Examine the sorted results. What percentile lands where? Plot a histogram. The shape tells you whether outcomes cluster or spread.

A 5,000-iteration run takes seconds on modern hardware; each passes through the same financial logic once, then discards its intermediate values.

Interpreting the output: ranges and tails

When the run finishes, you have a list of 5,000 valuation numbers. The arithmetic mean is one “fair value,” but the median, 25th percentile, and 75th percentile are often more useful. If the 25th percentile is $120 and the 75th is $180, half of your plausible scenarios fall in that band.

The shape of the distribution reveals risk. If it’s symmetric—bell-shaped—the mean and median align, and the business is “normally” uncertain. If it’s skewed left, the downside tail is long; a few bad scenarios crater the valuation dramatically. If skewed right, the upside is open-ended. This tail behaviour matters more than the central estimate if you’re a risk manager or contrarian investor.

The tails themselves are instructive. The 5th percentile tells you “barely one scenario in twenty falls below this.” The 95th, vice versa. Some practitioners widen these to 1st and 99th for stress-testing. If a 1st-percentile outcome is still solvent and profitable, the business is resilient.

When to use Monte Carlo, and when not to

Monte Carlo excels when assumptions carry material, quantifiable uncertainty and the distribution shapes are defensible. A biotech valuing a drug pipeline with success probabilities? Perfect. A mature utility projecting stable cash flows? Overkill.

The method also shines for screening. If a valuation is sensitive to only three variables—say, revenue growth, margin, and the discount rate—Monte Carlo quickly reveals which of your 5,000 scenarios survive a given hurdle price. That informs how tight your margin of safety needs to be.

The catch: garbage in, garbage out. If you define distributions carelessly—pulling percentiles from thin air rather than historical data or expert consensus—the output looks rigorous but is just random noise dressed as science. You must justify each distribution shape and parameter. A triangular distribution (most-likely, upside, downside) forces you to pick three points explicitly; a normal curve with a standard deviation requires thought about what “one standard deviation” means for your business.

Correlation assumptions matter too. Ignoring that revenue growth and margin often move together—or move opposite—can understate or overstate volatility. Professional simulations often include a correlation matrix to capture these linkages.

Practicalities: spreadsheets and dedicated tools

You can build a Monte Carlo model in Excel using RAND() and NORM.INV() functions to draw random samples, but 5,000 iterations become tedious. Most practitioners use dedicated software: @Risk, Crystal Ball, SimVoi, or Python libraries like scipy.stats and numpy. These let you define distributions with a few clicks, specify correlations, run iterations in parallel, and export percentile charts instantly.

The output—a valuation range with confidence intervals—then feeds into portfolio decisions. If you’re building a short-selling case, you care about the downside tail. If you’re a long-term compounder buying with a margin of safety, the median and 25th percentile might be your frame.

See also