Season Monte Carlo: the probability of meeting margin and profit
Plays the season many times with random demand and cost (with a fixed seed, the result is reproducible) and shows the probability of meeting the margin and profit targets, profit percentiles, the average tail of bad seasons, the stock-out probability and the expected leftover.
At a glance
- What question does it answer?
- With what probability the season meets the margin and profit targets and what a bad season looks like if demand and cost are random.
- What data is needed?
- Price, cost, leftover price, the buy and fixed costs.
- How to read the result?
- Example on the default values (10,000 runs, seed 2026): the probability of meeting a 15% margin is 96.8%, the profit target of RUB 3 million 87.1%, profit above zero 99.8%; average profit RUB 5.21 million, bad season P10 RUB 2.71 million. Example on the default values: Probability of meeting the margin target: 96.8%; Probability of meeting the profit target: 87.1%; Probability of profit above zero: 99.8%; Average profit: RUB 5.21 million. The figures are illustrative: enter your own to see your result.
Why and how to reason
An ordinary plan shows one figure: demand of 3,000 units, buying 3,300, profit of 5.89 million. But demand is random, and cost may fluctuate too. Monte Carlo plays out the season many times (10,000), each time with new random demand, and shows the distribution of results. A seed fixes the generator: the result is reproducible. For each run: sold = the lesser of buying and demand, leftovers sell at the leftover price, profit = revenue + leftovers × leftover price − cost × buying − fixed. Average profit (5.21 million) is below the planned one (5.89 million) because at high demand the buying cap cuts sales and at low demand unsold goods remain. The probability of meeting the profit target of 3 million is 87%, a bad season (P10) gives 2.7 million, and the average of the worst 5% of seasons 1.3 million.
When to use it
Reach for it for decisions with an irreversible buy to see the risk, not one average: how much you can lose and with what probability you will meet the target. Do not use it as a forecast: the result is only as good as your assumptions about the mean and spread of demand. Assumptions: the shape of demand is chosen (lognormal or normal), the spread of cost is independent of demand, enough runs.
How to read the result
Read probabilities and the tail, not the average. The probability of meeting the margin target (97%) and the profit target (87%) answers “will it be enough”. Percentiles P10, P50 and P90 (2.7, 5.6 and 7.1 million) show the range. The average profit of the worst 5% of seasons (1.3 million) is your “bad reality”: ask yourself whether you would survive it. The probability of shortage (31%) and average leftovers (500) show the balance between losses. Change the buy and watch how the probability of the target and the tail move: that is how to find a compromise between expectation and risk.
Common mistakes
Taking the demand spread too narrow (past seasons were “calm”), forgetting the spread of cost, believing the result as an exact prediction, not fixing the seed when comparing scenarios. To compare two buys use the same random numbers, otherwise the difference is noise. Remember that a fixed number of runs gives noise: for the final decision increase them to 20,000 and compare.
What data is needed?
- Price, cost, leftover price, the buy and fixed costs.
- Mean demand, demand spread and cost spread.
- Margin and profit targets, the number of runs and the seed.
Inputs
| Field | Unit | Value in the example | Meaning |
|---|---|---|---|
| Selling price excl. VAT | ₽ | 4,900 | The average selling price over full price and discounts. The example value is illustrative. |
| Unit cost | ₽ | 2,200 | Full cost of a unit at the warehouse. The example value is illustrative. |
| Leftover price | ₽ | 1,500 | What leftovers sell for at season end. |
| Buy | units | 3,300 | How many units you buy for the season: the simulation counts profit, leftover and shortage for it. |
| Season fixed costs | ₽ | 2,000,000 | Costs independent of sales: the collection team, brand advertising. |
| Mean demand | units | 3,000 | Expected sales over the season. Footwear, accessories and children’s have their own values: take the category data. |
| Demand spread (std. deviation) | units | 800 | An estimate from past seasons. The example value is illustrative. |
| Demand shape | Lognormal: never negative, a long upper tail | Lognormal suits when a hit season may bring a strong spike upward. | |
| Cost spread | % | 5 | Relative deviation of cost: exchange rate, material price. Your estimate. The example value is illustrative. |
| Link of demand and cost (correlation) | 0 | 0: independent, as before. Plus: in seasons with high demand the cost is higher (materials and rush orders get dearer); minus: the other way round. An estimate from past seasons. | |
| Spread of the selling price | % | 0 | 0: the price is fixed, as before. The relative deviation of the average price over the season (discounts, the price in a weak season). Your estimate. |
| Link of demand and price (correlation) | 0 | Plus: with high demand the price is higher (fewer discounts). Works when the price spread is above zero. | |
| Returns, mean share of sold units | % | 0 | 0: no returns, as before. Returned units go at the leftover price. |
| Spread of the return share | pp | 0 | 0: the return share is fixed. The standard deviation of the return share across seasons. |
| Link of demand and returns (correlation) | 0 | Plus: in seasons with high demand returns are higher. Works when the return spread is above zero. | |
| Demand spread from past seasons | No: set above | Takes the relative spread (forecast against actual) from the table below instead of the manual demand spread; at least three seasons are needed. | |
| Margin target | % of revenue | 15 | Profit ÷ revenue you want to get. Your target. |
| Profit target | ₽ | 3,000,000 | The cash result of the season you want to get (revenue minus buying and fixed costs; taxes and payment timing are not included). |
| Number of runs | 10,000 | More runs, less random error in the estimate. | |
| Generator seed | 2,026 | With the same seed the calculation gives the same numbers. Change the seed to see the estimate’s own scatter. |
How it is calculated
| Step | Formula |
|---|---|
| Demand | lognormal: exp(m + s·z) or normal max(0; μ + σ·z) |
| Season | sold = min(buy; demand); leftover = max(buy − demand; 0) |
| Profit | price × sold + leftover price × leftover − cost × multiplier × buy − fixed |
| Probability | the share of runs where margin or profit is not below the target |
Worked numbers
Default values from the app; in the interactive version you enter your own.
- Probability of meeting the margin target
- 96.8 %
- Probability of meeting the profit target
- 87.1 %
- Probability of profit above zero
- 99.8 %
- Average profit
- 5.21 m ₽
- Median profit (P50)
- 5.58 m ₽
- Bad season (P10)
- 2.71 m ₽In 10% of seasons profit is no higher than this.
- Good season (P90)
- 7.09 m ₽
- Average profit of the worst 5% of seasons
- 1.33 m ₽
- Average margin
- 35 %
- Stock-out probability
- 30.7 %
- Average leftover
- 500 units
- Shortfall below the average profit: VaR 95%
- 3.19 m ₽The average profit minus P5: by this much profit in the worst 5% of seasons is below the average. In roubles.
- Shortfall below the average profit: CVaR 95%
- 3.88 m ₽The average profit minus the average profit of the worst 5% of seasons: the expected shortfall in the bad tail. In roubles.
Data as a table
| Measure | Share of runs, % |
|---|---|
| 0.9…1.5 | 2.7 |
| 1.5…2.2 | 3.4 |
| 2.2…2.9 | 5.8 |
| 2.9…3.6 | 8.1 |
| 3.6…4.2 | 9.3 |
| 4.2…4.9 | 10.4 |
| 4.9…5.6 | 10.5 |
| 5.6…6.3 | 11.5 |
| 6.3…6.9 | 22.3 |
| 6.9…7.6 | 16.1 |
Data as a table
| Profit, m ₽ | Probability, % |
|---|---|
| 0.87 | 99 |
| 1.21 | 98.3 |
| 1.54 | 97.3 |
| 1.88 | 95.8 |
| 2.22 | 93.9 |
| 2.55 | 91.4 |
| 2.89 | 88.1 |
| 3.23 | 84.4 |
| 3.57 | 80 |
| 3.9 | 75.3 |
| 4.24 | 70.8 |
| 4.58 | 65.6 |
| 4.91 | 60.4 |
| 5.25 | 55.3 |
| 5.59 | 49.9 |
| 5.92 | 44.1 |
| 6.26 | 38.4 |
| 6.6 | 29.5 |
| 6.94 | 16.1 |
| 7.27 | 5.19 |
| 7.61 | 1 |
The result depends on your estimates of the mean and spread of demand: Monte Carlo adds no knowledge, it shows what follows from the assumptions. The seed is fixed so the numbers are reproducible; the random error of a probability is about one percent at 10,000 runs.
Taxes, payment timing, reordering and moving leftovers are not modelled. For footwear and children’s seasonal demand by size is more scattered: count by the size core.
Nuances and limits
- Trusting the probability more than the estimates of mean and spread it came from.
- Never changing the seed and not seeing how the estimate itself wobbles.
- Forgetting that reorders, taxes and payment timing do not take part in the calculation.