Backtesting a model portfolio: what a small RIA can learn from it, and what it can't tell you
A backtest shows how a model portfolio would have behaved over a past period if it had been held as designed. For a small RIA it's useful for comparing two versions of a model, seeing how deep past drawdowns went, and checking that a rebalancing rule behaves sensibly. It can't predict future returns, it ignores many real costs and timing, and it's easy to tune until the past looks good. Record why you ran it and what you decided, and treat any use of backtested results with clients as a compliance question for your consultant.
An advisor wants to move the Growth model from 70/30 to 80/20. They run a backtest over ten years, the 80/20 version comes out ahead, and the change goes through. Six months later, a client asks why the model changed. The answer, "it backtested better", is true, but it doesn't say much about the reasoning, and it leaves out everything the backtest couldn't show.
What is a backtest good for?
| Use | What it shows |
|---|---|
| Comparing two versions of a model | How the difference in weights played out over the same past period |
| Seeing past drawdowns | How far the model fell in bad markets, and how long it took to recover |
| Testing a rebalancing rule | Whether a calendar or drift rule behaved sensibly in practice |
| Explaining risk internally | A concrete picture of the ups and downs a client might live through |
What can't it tell you?
- The future. Past returns of a mix of assets say little about the next ten years.
- Real costs and timing. Trading costs, taxes, cash drag and the day you actually traded are often simplified or left out.
- Client behaviour. A drawdown on a chart feels different from one in a client's account.
- Whether you tuned it. If you tried twenty versions and kept the best, the winner is partly luck.
How should you record a backtest?
- Why you ran it: the question you were trying to answer.
- The period and the benchmark you compared against.
- The versions you tested, including the ones you rejected.
- What you concluded, and what else went into the decision.
- The model version that resulted, with its effective date.
Keep this with the model change itself, so the backtest is one input in a documented decision rather than the whole reason.
A worked example
Take a hypothetical two-person RIA reviewing its Growth model. The advisor backtests 70/30 and 80/20 over a period that includes a serious market fall. The 80/20 version shows higher returns and a deeper drawdown. The decision note says: clients in this model have long horizons and said in their reviews they could tolerate larger falls; the backtest showed the drawdown was deeper but recovered within the period tested; the change takes effect on a set date as version 4. The backtest stays internal.
What are the common traps?
- Choosing a start date that flatters the result.
- Testing many versions and remembering only the winner.
- Treating a backtest as a forecast.
- Putting backtested numbers in client material without checking the rules first.
When is this overkill?
For small adjustments within a model's existing range, a backtest adds little. Save it for changes that alter the model's character.
How Advorra fits
Advorra has a backtester alongside versioned model portfolios and their rebalancing rules, so you can test a proposed version and then record the change as a new version with its own date. See the product page, or read about versioning model portfolios and documenting investment decisions.
One thing to do this week: for your last model change, write down in two sentences why you made it. If the only reason is a backtest, add the rest.
A practical guide, not investment, legal or compliance advice. Ask your compliance consultant before using hypothetical performance with clients.
Questions advisors ask
Can I show backtested results to clients?
Hypothetical performance is subject to specific rules for advisers, and many firms keep backtests internal. Ask your compliance consultant before any backtest appears in client material.
How far back should a backtest go?
Long enough to include at least one serious market fall, so you see how the model behaved under stress. Note the period you used.
What is overfitting?
Adjusting a model until its past results look as good as possible. It usually makes the model look better in the past and no better in the future.
Should I backtest every model change?
For meaningful changes, such as a new asset class or a different weighting, it's a useful check. For small tweaks, it's often not worth the time.
Is this investment advice?
No. It is a practical guide to using backtests as one input in a documented decision.
See it on a demo firm
Advorra is practice software for independent RIAs: clients, models, IPS, billing and compliance records in one place, on top of the custodian you already use. Built by The CollabHub.
Get in touch