Optimization Is Backward-Looking. Good Advisors Are Forward-Looking.
An increasing number of wealth management firms are selling something that sounds genuinely useful: an algorithm that looks at your financial goals, runs thousands of calculations, and gives you the “optimal” portfolio. Goals-based. Multi-period. Personalized. The marketing is sharp, and the math isn’t wrong.
It’s also not enough.
Portfolio optimization is a tool we use every day. But anyone pitching the optimizer as the answer is telling you something important about what they don’t plan to do next. Because the one thing an optimization engine cannot do is see forward.
What optimization actually is
The math works like this. You feed a model a set of asset classes (stocks, bonds, cash, sometimes alternatives), along with estimates of how much return each will produce, how much they’ll bounce around, and how they move in relation to each other. The algorithm then solves for the mix of asset classes that gives you the highest expected return for a given level of risk, or the lowest risk for a given expected return. You get an “efficient frontier” of portfolios. Pick your preferred point on the curve, and that’s your allocation.
This is real, useful work. Done well, with thoughtful return assumptions, realistic volatilities, and sensible constraints, it produces sensible portfolios. Done carelessly, it produces portfolios that look clever and fail in ways that only show up later.
But here’s the essential limitation: every input into the optimizer comes from the past. Expected returns are typically anchored to long-run historical averages, sometimes adjusted for current valuations. Volatilities and correlations come from historical data. The machine can solve for the optimal portfolio assuming the next decade looks something like the last several. When the regime changes, the optimizer doesn’t know.
When the regime changes
In 2022, the traditional 60/40 portfolio had its worst year since the 1930s. Stocks fell roughly 18%. Bonds fell roughly 13%. The blended portfolio lost roughly 16%, not because the diversifier was broken, but because both were reacting to the same driver: a rate shock from rising inflation. The negative correlation between stocks and bonds that the textbooks depended on, the assumption embedded in every balanced-fund optimization since the 1980s, stopped working.
An optimization engine trained on the forty years before 2022 would have told you bonds were a reliable hedge for equity risk. The math said so. The covariance matrix said so. The historical data said so. And it was wrong.
No algorithm flagged the regime change in advance. The advisors who were thinking about fiscal dominance, the end of disinflation, and the positive stock-bond correlation that had always accompanied inflationary regimes had already adjusted. They didn’t need the optimizer to tell them. The optimizer would have been the last to know.
What the advisor brings
The optimizer sees history. The advisor brings the present and the future, not as prediction, but as framework. A good advisor is doing things no algorithm is doing:
-
Identifying the regime we’re actually in. Is inflation structural or cyclical? Is the Fed still the dominant force in markets, or is fiscal policy? Is this a growth-deflating world, a growth-inflating world, a stagflationary world? The answer drives allocation.
-
Recognizing when the past is a poor guide to the future. An optimizer using forty years of disinflationary data to allocate between stocks and bonds in a fiscal-dominance regime is optimizing the wrong problem.
-
Exercising judgment about what belongs in a portfolio that an algorithm wouldn’t select on its own. Managed futures that hedge against regime changes but lose money in calm markets. Energy equities at a weight above any index’s. Gold exposure held inside a trend-following strategy rather than a static allocation.
-
Knowing when to act on the judgment. The hardest part. A thesis change is useless unless you reposition when the evidence says so, and hold the position when the market hasn’t priced the change yet.
The starting point is the return you need
There’s one thing the algorithmic platforms get right that traditional wealth management often misses: the starting point should be the return the plan actually needs, not a risk-tolerance questionnaire.
A 68-year-old couple who needs 4.5% after inflation to fund a comfortable retirement is in a different position than a 45-year-old who needs 7% to hit a target net worth by 65. The allocation for each one falls out of the math of their goal, not out of a 1-to-10 slider.
We use this anchor. We run the financial plan, surface the required return, and design portfolios against it. If the required return is 5% and we can build a reasonable portfolio to target it with moderate risk, we do. If the required return is 9% and the honest read is that no portfolio gets there reliably without unacceptable risk, we say that, and the conversation pivots to spending, saving, or timing, because the portfolio isn’t the problem to solve.
What we don’t do is stop at the optimizer’s answer. The required return sets the target. The optimizer proposes a mix. Then the real work begins.
The test
If you’re being shown an optimization-driven wealth platform, ask this question: When did your model change its mind about energy? About bonds? About inflation? About emerging markets?
Those aren’t model outputs. Those are decisions made by humans looking at the evidence and updating their framework. If the answer is “the model adjusts automatically based on forward-looking asset class returns,” follow up with: Who decides those returns, and how often do they change? Usually it’s a research team at the parent firm, updating quarterly or annually. That’s one group of humans, one cadence, and one framework, all embedded in the platform and invisible to you.
We prefer a different arrangement. We read multiple research sources, we debate the signals, we keep our own regime framework, and we move when the evidence says to move. Some of it is encoded in tools. Most of it is judgment. None of it is automated.
What we believe
Optimization is a tool worth using. It disciplines the process, surfaces the tradeoffs, and prevents obvious mistakes. We would not build a portfolio without running the math.
And we would not build a portfolio only by running the math. The machine can tell you the historically optimal mix. It cannot tell you when history is a bad guide. That’s what the advisor is for. If the advisor is reduced to a front-end for an algorithm, the client has lost the most valuable thing a real advisor provides: someone actually thinking, in real time, about whether the model is right.
Models describe the past. We’re paid to think about the future.
If your current plan came out of a model and no one has ever shown you the judgment behind it, ask to see ours. The consultation is a conversation, not a sales call.
This article reflects the views of Long Point Wealth Management as of the date written. It is intended for educational purposes and should not be construed as personalized tax, legal, or investment advice. All investing involves risk, including loss of principal. Past performance does not guarantee future results.