Backtest: what two backtests show and what they don't
A backtest applies a fixed rule to past prices. It shows what would have been, not what is coming. This chapter has two parts: a portfolio that selects by grade, and the question of whether a particular weighting of the components would have beaten the MSCI World.
Important: Both are hypothetical backtests. They are not investment advice and not a promise for the future.
Part A: Portfolio by grade
Backtest over 3 years (30 Oct 2023 to 30 Oct 2026) on the entire universe, leverage 1, weekly update. A stock is only replaced when a new one is at least 20 points better.
Conclusion: In the backtest, the portfolio would have turned €10,000 into €22,101, which is €3,706 more than the MSCI World. The largest decline was −16 % (MSCI World −17 %).
1. The rules
- Basis: 3,111 stocks, of which 1,776 with a fundamentals score can be bought.
- Portfolio: at most 20 stocks, Long only, leverage 1, €500 per slot at €10,000.
- Selection: the stocks with the highest grade, at least +15.
- Update: every week (Friday close, trading on Monday).
- Replacement: a new stock replaces the weakest one in the portfolio if its grade is at least 20 points higher (lead 20).
- Stop-loss: 25 % below the purchase price. Take-profit: 20 % above it, each checked intraday. Slots that become free stay in cash until the next update.
- Costs: 0.05 % spread per side, no taxes, no currency effects.
2. Results at a glance
| Top 20 portfolio (lead 20) | MSCI World | Stock basket, held | |
|---|---|---|---|
| €1,000 becomes | 2,210 € | 1,839 € | 2,056 € |
| €10,000 becomes | 22,101 € | 18,395 € | 20,557 € |
| Result | +121 % | +84 % | +106 % |
| per year | +30 % | +23 % | +27 % |
| largest decline | −16 % | −17 % | −26 % |
With €10,000 that is €500 per slot at the start, with €1,000 it is €50. Fractions of shares are allowed. The annual return of the portfolio (+30 %) is calculated from the final value over exactly three years.
For comparison: The stock basket consists of the same buyable stocks, equally weighted and simply held. It shows what would have come out with no selection at all. The portfolio was 15 percentage points above it.
3. How much does the result depend on the settings?
Lead required for a replacement (take-profit 20 %)
| Lead | Result | €10,000 becomes | Largest decline | Buys / sells | Replacements |
|---|---|---|---|---|---|
| 0 | +77 % | 17,672 € | −18 % | 1,040 / 1,021 | 821 |
| 5 | +70 % | 16,965 € | −19 % | 767 / 748 | 544 |
| 10 | +79 % | 17,934 € | −19 % | 638 / 619 | 402 |
| 15 | +111 % | 21,071 € | −16 % | 503 / 485 | 265 |
| 20 | +121 % | 22,101 € | −16 % | 403 / 386 | 172 |
| 30 | +116 % | 21,624 € | −16 % | 334 / 317 | 100 |
From about 14 points upwards, the result is consistently +100 to +120 %; below that it is +70 to +86 %. Fewer changes helped in this period. The MSCI World is at +84 %.
Part B: The weighting M01118 in the component test
In short: M01118 reweights the components of the dashboard rating; the exact weights are not published. In the development years (2013–2017 and 2023–2026) it achieved an annual return of 22.6 % versus 15.4 % for the MSCI World, with a Sharpe of 1.16 versus 1.09 and a maximum loss of −25.7 % versus −23.2 %.
4. What is M01118?
The dashboard combines several components, each with a weight, into one grade. The component test asks: would a selection based on this grade have beaten the MSCI World? M01118 is one of 10,003 weightings tested. These components are tested:
| Component | Meaning |
|---|---|
| Fundamentals | Annual financial statements, 8 criteria |
| Trend | Trend direction |
| Verdict | overall technical check |
| SMA50 | Position relative to the 50-day average |
| Strength | Trend strength |
| RSI | RSI |
| Maturity | How mature the trend is |
| Headwind | Counter candle, earnings date (shortened) |
| Rank | Room to the upside (distance to the resistance zone) |
| Relative strength | 12–1 versus SPY |
| B-Score | backtested approximation |
| Rate | backtested approximation |
What the most important components measure
- Fundamentals: Score of 0–8 from eight criteria of the annual financial statements (SEC 10-K), using only reports that had already been published on the signal day.
- Relative strength 12–1: Return of the stock from 365 days ago to 28 days ago (the last month is left out) minus the same return of SPY, divided by 20 % and capped at −1 to +1. If the stock had no price twelve months ago, the component is unknown. The formula is fixed and has not been adjusted.
- B-Score: Buffett score, in the backtest only reconstructed as an approximation from the 10-K and the prices.
- Technical components (Verdict, SMA50, Strength, Maturity), Headwind and Room to the upside each contribute only a little.
Trading rules of the test (fixed, not part of the search)
- Every week, the 20 stocks with the best grade are bought, equally weighted, Long only, without leverage, minimum grade above 0; purchase at the open of the following day. A stock is dropped when it falls out of the best 40. The rest stays in cash (without interest).
- Universe: Members of the S&P 500 on the respective day, only stocks with traceable Fundamentals, Headwind and Room to the upside. Stocks that have left the index are missing.
- Benchmark: iShares Core MSCI World (accumulating), in euros. Result in euros, after costs.
5. How M01118 was selected
10,003 configurations were calculated in the development years (6,000 widely spread, 4,000 close to the dashboard weights, plus comparison rows). Together with the earlier search, that makes 37,625 trials on the same data.
| Metric across all 10,003 configurations | Value |
|---|---|
| Median Sharpe | 0.99 (index 1.09) |
| Share with Sharpe of at least 1.09 | 22.5 % |
| Share with CAGR above 15.4 % | 63.3 % |
| Highest Sharpe / highest CAGR | 1.29 / 32.7 % |
| Plan points 1–4 met in the development years | 1,464 configurations |
| … of which with a small Rate share | 687 configurations |
Selection rule (fixed before looking at the selection)
- Step 1: Plan points 1 to 4 met in the development years: CAGR above the index, Sharpe at least 1.09, maximum loss at most 1.25 times that of the index (−29.0 %), above the index in at least 60 % of calendar years.
- Step 2: The rate component may only have a small share. The coefficients of the rate model were fitted on stocks that also fall within the test years (a data leak in favour of the rate). Without the rate, the best configuration still reaches a Sharpe of 1.20, so the rate is not needed.
- Step 3: Ranking by the smaller of two values: the weighting's own Sharpe and the average Sharpe of the 20 most similar weightings. An outlier whose neighbours are poor fails this way.
Result: M01118 is at the top of this ranking. By pure Sharpe it is at rank 437 of 10,003 (94th percentile), so it is deliberately not the front runner. The front runner (Sharpe 1.26) is less reliable because of the selection effect with so many attempts. The neighbourhood of M01118 is stable: its 20 most similar weightings have an average Sharpe of 1.14.
6. Result in the development years
Calculated on 2013–2017 and 2023–2026, after costs, in euros. The configuration was selected in these years, so the figures tend to be too good.
| Metric | M01118 | MSCI World | Meaning |
|---|---|---|---|
| CAGR | 22.6 % | 15.4 % | Annual return, geometric average |
| Sharpe ratio | 1.16 | 1.09 | Return per unit of volatility; higher is better |
| Maximum loss | −25.7 % | −23.2 % | largest decline from the peak |
| Years above index | 89 % (8 of 9) | – | Share of calendar years with excess return |
| Turnover per week | 7 % | – | Share of the portfolio that is swapped each week; low means low costs |
| Adjusted Sharpe (DSR) | 0.62 | – | factors in the 37,625 trials; below 0.5, success can hardly be told apart from chance |
| Neighbour Sharpe (20 most similar) | 1.14 | – | Stability of the surroundings |
How to read this: The return is clearly above the index, but the protection against declines is no better (−25.7 % versus −23.2 %). At 0.07, the Sharpe lead is small. The corrected Sharpe of 0.62 is above the mark of 0.5, but it is not strong evidence.
Terms
| CAGR | average annual return (geometric), after costs, in euros |
| Sharpe ratio | Excess return per unit of volatility; compares results with different levels of risk |
| Maximum loss | largest decline from the peak to the following low |
| Years above index | Share of calendar years in which the return was above that of the MSCI World |
| Turnover per week | Share of the portfolio that is replaced per week |
| DSR (deflated Sharpe) | Sharpe ratio that factors in the number of trials; close to 1 means hardly any chance, below 0.5 hardly distinguishable from chance |
| Development years | 2013–2017 and 2023–2026: this is where the search and the selection took place |
Hypothetical backtests, not investment advice. Past results say nothing about future ones.
Education, not investment advice. · Next: The six Buffett criteria · ETF or single stocks? · Glossary