Bottom-Up Valuation Model
Value a business line by line so forecast errors cancel instead of compound.
A bottom-up valuation model values a company by reconstructing each individual revenue stream, cost line, and operating driver instead of slapping a single growth rate on top-line revenue. For Tesla, that means modeling each factory, each production line, the cars produced per year, regional margins, and unit economics, then aggregating up. The point is not perfection but reducing the magnitude of error so your target price falls inside a reasonable range.
Top-down models multiply errors: a 5pp swing in growth rate compounds over ten years into wildly different valuations. Bottom-up models let independent errors cancel — being slightly high on wages in one geography offsets being slightly low on output in another. The output is a fair-value estimate you would actually pay today, not a price prediction for some future date.
The model is alive: every quarter you re-read filings, listen to earnings calls, update assumptions, and shift the valuation as the company and competitive landscape evolve. The model is the discipline; the number falls out of the discipline.
- You are buying a fractional share of a business, so value it the way a buyer of the whole business would.
- Bottom-up modeling reduces the magnitude of error because independent line errors tend to cancel.
- Top-down growth-rate models multiply errors and can produce 2-3x swings in target price.
- Your valuation is a present-value fair price, not a future stock-price prediction.
- The valuation is a living number that updates every quarter as new information arrives.
- Filter to one company worth modelingScreen the universe of stocks down to a single company you actually want to value. Use whatever filters fit your style — sector, size, profitability, conviction. Most companies you analyze you will never invest in.Pro tipSpend more time on companies you reject than on the ones you hold; the reject pile sharpens your eye.
- Read every filing and callConsume the 10-Qs and 10-Ks (the long versions filed with the SEC, not the social-media summaries), all official news, appointments, and the earnings call transcript. Then do the same for the main competitors.Pro tipThe long 10-K version often contains lines that never appear in the press release — that is where the alpha hides.
- Decompose the business into its driversBreak the company into individual business lines, revenue streams, factories or production lines. For each, identify the unit drivers: price per unit, volume, capacity, regional margins, marketing efficiency.
- Build the bottom-up model line by lineForecast each driver individually rather than applying a blanket growth rate. Aggregate revenue lines, attach their cost bases, layer in shared corporate costs and marketing spend.WarningResist the urge to set a 30% growth rate across the board — that is exactly the top-down trap that compounds error.
- Run scenarios as probability-weighted overlaysLayer scenarios on top of the base model: technological jumps, regulatory shifts, macro events. Assign probabilities, account for correlations between scenarios, and run a Monte Carlo-style simulation across the distribution.Pro tipARK Invest publishes their Monte Carlo distributions — useful as a reference for how to present scenario ranges.
- Derive a fair-value-today priceOutput is a present-value range — for example, an 80-120 fair-value range with the spike around 100. Compare to today's share price; the gap is your upside or your warning.
- Update every quarter foreverEach quarter, re-read all filings (~7-10 hours per company), listen to the earnings call (~2 hours), and adjust the model (~2-3 hours). Valuations shift materially over years as the company and competition evolve.WarningBudget at least 200 hours per quarter for a 10-position portfolio; if you don't have that time, you are gambling, not investing.
Sasha's Tesla model breaks down individual factories and production lines, estimates cars per line per year, and compares regional margins (China vs Texas vs hypothetical Brazil/Indonesia/India sites), including unionization risk in Germany. He layers scenario overlays for productivity jumps like robotic production lines.
Sasha held AMD for years through a long run-up, loving the company's products and Lisa Su's leadership. When the share price reached his model's fair value he sold despite being a fan, because there was no upside left in his valuation.
Sasha trained as a mathematician at Oxford and spent years inside large American banks building valuation models, then ran a strategy consultancy doing M&A valuations for banks acquiring smaller financial firms. That work — pricing what a buyer should pay for an entire business — translates directly into how he values listed companies as a fractional owner.
He contrasts this with the social-media investing crowd that treats stocks like a fan club, and built his YouTube channel in 2020 specifically to share the analysis process he saw missing in retail finance content.