Scenarios and risk: tornado, season Monte Carlo, cash gap probability, VaR-lite, stress tests
A quantitative view of season uncertainty: which assumptions move profit most (tornado), what the spread of the result is (Monte Carlo), with what probability the season turns loss-making or hits a cash gap, which “bad case reserve” is needed (VaR-lite) and what happens under hard stress. The decision: purchase size, the cash cushion and plan revision triggers.
Why this document
The finance director and the owner read this when approving the season budget and deciding what buffer to hold. A single profit forecast hides uncertainty, yet decisions are made precisely under it. A tornado chart shows which assumption moves the result most and what to watch, Monte Carlo gives the probability of a loss-making season or a cash gap. A stress test answers whether the business survives a bad but realistic scenario. Without this, the buffer and the credit limit are chosen at random.
At a glance
- What is this document and why is it needed?
- A quantitative view of season uncertainty: which assumptions move profit most (tornado), what the spread of the result is (Monte Carlo), with what probability the season turns loss-making or hits a cash gap, which “bad case reserve” is needed (VaR-lite) and what happens under hard stress. The decision: purchase size, the cash cushion and plan revision triggers.
- Who owns the document and how often is it rebuilt?
- Finance director and planning analyst; calculated before approving the season budget and at every noticeable change in FX, demand or terms; the result goes to the board and treasury limits.
- What mistakes are the most common?
- A single “pessimistic” scenario is built → probability and size of losses unknown → tornado, Monte Carlo and stress together. Risk is measured on profit only → a profitable season runs into a cash gap → a separate monthly cash model. Independent drivers are set → the tail is understated → correlation of demand and markdown plus an “everything bad” stress.
For whom
- Brand
- A brand bears purchase risk in advance: the volume is fixed months ahead while demand, FX and partner returns are unknown; the main drivers are season demand, purchase FX, markdown and partner returns. A cash gap arises from the gap between production payment and wholesale receipts.
- Retail
- A retailer bears risk of leftover stock and fixed costs (rent, staff): when traffic falls costs do not, and the purchase is already paid. The main drivers are traffic and conversion, markdown, rent and purchase FX; a cash gap falls on the peak of supplier payments before the high season.
What it consists of
The season model
- Illustrative model: purchase of 10,000 units at a cost of RUB 1,000, price RUB 3,300 excluding VAT, mean demand of 9,400 units at full price, returns of 6%, average markdown 12%, leftovers sold at 30% of price, fixed costs of RUB 10.5 million. Base season profit is RUB 5.75 million.
- The model must be simple: 5-8 drivers that the business really discusses; the more parameters, the less trust in the result and the harder it is to explain.
Tornado: what moves profit
- Each driver is moved to the bad and good side with the others at base and the profit change is read. The strongest driver in the example is Season demand ±15% (from -2.45 to +RUB 1.04 million); next is Markdown depth 8% / 18% (from -1.75 to +1.17).
- Tornado asymmetry is information: demand is capped by the purchase, so demand above plan adds almost no profit while a fall hits in full. The chart shows where to hold reserves and what to watch weekly.
Monte Carlo: the distribution of the result
- Drivers are given distributions (demand lognormal with σ 18%, markdown normal ±3.5 pp, returns ±1.5 pp, purchase FX σ 5%, fixed costs σ 5%) and 10,000 seasons are simulated with a fixed seed. Mean profit is RUB 4.96 million, below the base because demand is capped by the purchase; the probability of a loss is 2.1%.
- Correlations matter: weak demand usually comes with deeper markdowns; an independent model understates the tail. If the link is known it is set by a coefficient; if not, a stress scenario where everything is bad at once is added.
Cash gap, VaR-lite and stress
- Cash by month: opening balance RUB 10.2 million, purchase paid 50/50 in the first two months, sales along the season curve. The probability that the balance dips below zero at least once is 13.8% although the season loses money in only 2.1% of cases: a profitable season can run into a cash gap.
- VaR-lite: the 5th percentile of profit is RUB 0.86 million (in 95% of cases the result is no worse), the mean of the worst 5% (expected shortfall) is −RUB 0.05 million. This is a simple estimate, not a formal financial-market VaR: the distribution is modeled by assumptions.
- Stress tests: demand −25% gives RUB 1.67 million, FX +20% gives 3.75, both together -0.33, and with a deep markdown −RUB 3.18 million. For each stress actions are set in advance: cut the reorder, speed up markdown, use the credit line.
Brand and retailer: what the risk model contains
- Brand: brand drivers: partner demand (sell-out), purchase FX, returns and compensation to partners, factory lead times and production payment; the key cash gap is the time from production payment to wholesale receipts.
- Retailer: retailer drivers: traffic and conversion, average ticket, markdowns, rent and staff, purchase FX; the key cash gap is the peak of supplier payments before the high season while fixed costs run.
- Brand: brand stress: a factory delivery failure, loss of a partner, FX up 20%, partners refusing the season order; actions on orders and financing are set in advance for each.
- Retailer: retailer stress: traffic falling 15-20%, a mall closure, a delivery failure, FX and rent rate growth; actions on purchasing, space and staff are set in advance for each.
- Brand: the brand tornado shows that the purchase volume limits the gain from demand: flexibility (fabric reorders, small lots) and partner return terms matter for a brand.
- Retailer: the retailer tornado shows the power of fixed costs: with weak traffic rent and staff do not fall, so variable rent parts and schedule flexibility matter.
- Brand: a brand closes a cash gap with partner advances, a credit line, factoring or supplier deferral; terms are agreed before the season, not at the moment of the gap.
- Retailer: a retailer closes a cash gap with a credit line, supplier deferral, faster markdown of leftovers and shifting capital expenditure; the action plan is tied to cash balance thresholds.
- Brand: a brand shows the board VaR-lite together with the cash gap probability: the season purchase size is decided on these two numbers, not on base profit.
- Retailer: a retailer shows VaR-lite together with covenant headroom: if covenants are breached in a bad season the purchase is cut or agreed in advance with the bank.
Where the data comes from
| Data source | Fields |
|---|---|
| Season budget | Sales plan, price, cost, returns, markdowns, fixed costs, payment and receipt schedule. |
| History of deviations | Deviations of actual from plan in demand, markdown, returns and FX over past seasons: driver spreads are taken from them. |
| Treasury | Cash balances, credit lines and limits, covenants, overdraft rate and terms. |
How to build it step by step
- Step 1. Build the season profit and cash model on 5-8 drivers; reconcile the base with the approved budget.
- Brand: Step 2. Add the gap between production payment and wholesale receipts and partner returns as separate drivers.
- Retailer: Step 3. Add traffic and conversion, rent and fixed costs and the peak of supplier payments before the season.
- Step 4. Set driver ranges (bad, base, good) from the deviation history, not by feel.
- Step 5. Build the tornado: shift each driver with the others at base; sort by range.
- Step 6. Set driver distributions and correlations, simulate at least 5,000 seasons with a fixed seed.
- Step 7. Obtain the mean, percentiles, loss probability, VaR-lite and expected shortfall; check stability against the number of runs.
- Step 8. Build the monthly cash flow for every run and find the probability of going negative and the percentiles of balance by month.
- Step 9. Set 3-5 stress scenarios with actions and triggers and compute the result of each.
- Step 10. Determine the cash cushion and purchase volume: the cushion covers the worst month at the percentile level, the purchase is cut if the gap probability exceeds the allowed one.
Formulas
| Measure | Formulas | Note |
|---|---|---|
| Season profit | Π = min(Q, D)·(1 − r)·P·(1 − m) + (Q − min(Q, D))·P·k − Q·C·φ − F | Q purchase, D demand, r returns, m markdown, k liquidation price, φ FX multiplier, F fixed costs. Assumption: returned units are not resold and do not enter stock, their cost is already paid, so the estimate is conservative. If returns can be resold at the liquidation price, r·min(Q, D)·P·k is added to profit: about RUB 0.56 million in the example. |
| Tornado range | R_j = max(|Π(x_j^low) − Π₀|, |Π(x_j^high) − Π₀|) | Drivers are sorted by R_j. |
| Monte Carlo estimate | P̂(Π < 0) = (1/N)·Σ 1[Π_i < 0]; error ≈ √( p̂(1 − p̂)/N ) | For N = 10,000 and p = 2% the error is about 0.14 pp. |
| VaR-lite and expected shortfall | VaR₅ = P₅(Π); ES₅ = mean of Π over the worst 5% of runs | Not a regulatory VaR; it rests on the assumed distributions. |
| Cash balance | S_t = S_(t−1) + Receipts_t − Purchases_t − Fixed_t; gap if min_t S_t < 0 | Receipts depend on the season curve and deferrals; payments on the supplier schedule. |
| Lognormal demand | D = D₀·exp(σ·Z − σ²/2), Z ~ N(0, 1) | The −σ²/2 correction preserves the mean D₀. |
| Cash cushion | Cushion = max(0, −P₁₀(min_t S_t)) | The percentile level is set by risk policy; 10% is illustrative. |
Formulas in notation
- Demand
D = D₀·exp(σ·Z − σ²/2), Z ~ N(0; 1)
Lognormal demand preserving the mean.- Loss probability
p̂ = (1/N) · Σ_i 1[ Π_i < 0 ]
Monte Carlo estimate; error √(p̂(1−p̂)/N).- VaR-lite
VaR_5 = P_5(Π); ES_5 = E[ Π | Π ≤ P_5(Π) ]
Quantile and tail mean.
Worked example
Season stress tests: profit under driver shifts, RUB million
| Scenario | Profit, RUB million | Change vs base, RUB million |
|---|---|---|
| Base | 5.75 | 0 |
| Demand −25% | 1.67 | −4.09 |
| FX +20% | 3.75 | −2 |
| Demand −25% and FX +20% | −0.33 | −6.09 |
| Demand −25%, FX +20%, markdown 25% | −3.18 | −8.93 |
Base profit RUB 5.75 million; Monte Carlo (10,000 runs): mean 4.96, 5% VaR-lite 0.86, loss probability 2.1%, cash gap probability 13.8%. The “demand −25%, FX +20%, markdown 25%” scenario is loss-making: actions to cut the purchase and use the credit line are needed. All inputs are illustrative.
Figures
- Bad shift
- Good shift
Data as a table
| Measure | Bad shift | Good shift |
|---|---|---|
| Season demand ±15% | −2.45 | 1.04 |
| Markdown depth 8% / 18% | −1.75 | 1.17 |
| Fixed costs ±10% | −1.05 | 1.05 |
| Purchase FX ±10% | −1 | 1 |
| Returns 9% / 3% | −0.82 | 0.82 |
Tornado: profit change under driver shifts. Each row shows the bad and good shift; demand hits down harder than it helps up because it is capped by the purchase.
Data as a table
| Measure | Share of runs, % |
|---|---|
| -2 | 0.7 |
| -1 | 2.3 |
| 0 | 5.6 |
| 2 | 9.7 |
| 3 | 14.3 |
| 4 | 17.7 |
| 5 | 20.4 |
| 6 | 17.2 |
| 8 | 9.2 |
| 9 | 2.5 |
Season profit distribution (Monte Carlo). The distribution is skewed left (demand is capped by the purchase): the left tail is longer; loss-making runs are 2.1%. Axis: lower bound of the profit interval, RUB million.
- Percentile 10
- Median
- Percentile 90
- Zero
Data as a table
| Measure | Percentile 10 | Median | Percentile 90 | Zero |
|---|---|---|---|---|
| 1 | 4.27 | 4.67 | 5.06 | 0 |
| 2 | −0.17 | 0.88 | 1.79 | 0 |
| 3 | 2.57 | 4.65 | 6.11 | 0 |
| 4 | 5.83 | 9.18 | 11.3 | 0 |
| 5 | 8.46 | 13 | 15.7 | 0 |
| 6 | 12 | 15.4 | 18 | 0 |
Cash balance by month: the 10-50-90% corridor. The lower bound of the corridor goes negative in the second month when the purchase payment is done and sales are still small. Axis: season month.
Data as a table
| Measure | Profit, RUB million |
|---|---|
| Base | 5.75 |
| Demand −25% | 1.67 |
| FX +20% | 3.75 |
| Demand −25% and FX +20% | −0.33 |
| Demand −25%, FX +20%, markdown 25% | −3.18 |
Stress tests: season profit. Demand and FX separately keep profit positive, together they take it to about zero and with a deep markdown into a loss.
Specifics by category
- Apparel
- The main risk is weather and the season purchase volume; demand is capped by the purchase so the tornado is asymmetric; the cash gap falls on production payment before sales start.
- Footwear
- A long cycle and imports: the purchase FX and size risk matter most; leftovers by size liquidate worse so the liquidation price in the model is lower and the loss tail heavier.
- Accessories
- The gift sales peak concentrates risk in a few weeks: a weekly model and a “peak failed” stress; turnover is high, the gap short but deep.
- Kids
- The school season with a hard date: a stress on delivery failure; size leftovers are lost fast so the liquidation price is low; demand depends on demography and family income.
- Men
- Demand is steadier, demand σ lower, the loss tail lighter; basics allow carrying leftovers so the liquidation price is higher; the main risks are FX and the prices of cotton and materials.
- Women
- A wide fashion range: demand σ is higher, markdowns deeper and correlated with weak demand; in the model the positive correlation of demand and markdown matters.
Common mistakes
- A single “pessimistic” scenario is built → probability and size of losses unknown → tornado, Monte Carlo and stress together.
- Risk is measured on profit only → a profitable season runs into a cash gap → a separate monthly cash model.
- Independent drivers are set → the tail is understated → correlation of demand and markdown plus an “everything bad” stress.
- Driver spreads are plucked from the air → a pretty but empty figure → spreads from the deviation history.
- VaR-lite is taken for a guarantee → the worst case can be worse → show expected shortfall and stress with actions.
- Actions under stress are not named → a number but no decision → each stress gets an action, an owner and a deadline.
Russian specifics
- Credit lines and loans often contain covenants (DSCR, debt/EBITDA, minimum balance): the stress test checks whether covenants would be breached in a bad season and agrees a waiver with the bank in advance.
- The payment schedule to suppliers and taxes (VAT, income tax, contributions) sets the calendar of cash outflow; in Russia tax payments fall on fixed dates and must be in the cash model. Rule: Tax Code Art. 164: Tax Code, Art. 164 (VAT rates) (rule check not completed).
- The provision for inventory write-down in a stress scenario must be assessed under FSBU 5/2019: an impairment loss on inventory reduces profit and covenant ratios. Rule: Order 180n of 15.11.2019: Ministry of Finance Order of 15.11.2019 No. 180n (FSBU 5/2019 Inventories) (rule check not completed).