A portfolio can contain dozens of holdings and still make one large bet. Different names may rely on the same customers, financing conditions, geography, or market narrative. To understand diversification, move beyond the count of positions and inspect their weights and relationships. This guide uses Modic's existing two-exposure model and the new rebalancing game to make those relationships concrete without recommending particular investments.
What is portfolio diversification?
Diversification spreads exposure across holdings whose outcomes are not identical. It can reduce the influence of a single adverse outcome and, under a specified model, reduce aggregate variability. It cannot guarantee a gain or eliminate every shared shock. Different investments can still respond similarly to inflation, funding pressure, or a broad change in risk appetite.
The SEC's asset-allocation guidance distinguishes spreading holdings from selecting an overall asset mix. Those are related decisions, but neither is determined by counting positions. A portfolio's purpose, horizon, liquidity needs, and risk tolerance matter. This article explains the model mechanics rather than providing a suitable allocation for an individual reader.
Why correlation matters more than a list of names
Correlation measures how two quantities tend to vary together, under a specified measurement period and definition. In the diversification experiment, both hypothetical exposures have standard deviation ten. The correlation control changes how much their fluctuations offset or reinforce each other. It does not change a company's quality, expected return, or valuation; those features are deliberately outside this model.
A zero correlation does not mean that two real assets can never fall together. A positive correlation does not mean they move identically every day. Historical estimates depend on the data window, and relationships can change. Use the model to understand a conditional relationship, then ask how reliable the estimated inputs would be before using the result as a description of a real portfolio.
A worked example: two equally variable exposures
Take two invented exposures with standard deviation ten each and put half the portfolio in each. The formula combines their variances and covariance. These units are abstract model variability, not forecast annual percentage returns or a recommendation to use a 50/50 investment mix. The equal inputs keep the effect of correlation easy to inspect.
- Set the share in A to 50% and correlation to zero. Combined standard deviation is 10×√0.5, about 7.07 units.
- Keep the weights unchanged and raise correlation to +1. Combined variability becomes ten: splitting between perfectly co-moving exposures produces no offset in this model.
- Set correlation to −1. Equal opposite fluctuations cancel in this idealised pair, giving zero variability; exact persistent cancellation is a strong assumption, not a realistic guarantee to pursue from this game.
The portfolio formula tells you which inputs matter
For weight w in A, weight 1−w in B, equal standard deviation σ, and correlation ρ, the model's standard deviation is σ×√(w²+(1−w)²+2ρw(1−w)). Weights and correlation both matter. A portfolio mostly concentrated in A behaves differently from an equal split even when the holdings are the same. Unequal real-world asset volatilities require the more general covariance formula.
Markowitz's Nobel lecture provides foundational portfolio-theory context. The interactive model is intentionally smaller than a portfolio optimisation problem. It does not estimate expected returns or select an efficient frontier. Its value is letting you see one relationship clearly before dealing with the uncertainty of inputs and the competing objectives of an actual investment plan.
Go a little deeper: Different variability in the two holdings
The general two-asset variance is w²σA²+(1−w)²σB²+2w(1−w)ρσAσB. Taking its square root gives standard deviation. An apparently balanced money allocation can still have unbalanced variability contributions when σA and σB differ.
Look for shared mechanisms and outsized weights
Write down the biggest positions and the mechanisms each depends on. A broad-sounding set may still be concentrated in one industry, customer base, country, or financing environment. You do not need to calculate a precise correlation to identify that all the holdings need the same favourable event. This qualitative map complements, rather than replaces, the quantitative measurement.
The Pareto concentration experiment helps separate the number of contributors from the share dominated by the largest ones. A portfolio's exposure can be concentrated even when the small tail contains many names. Ask how a loss in the largest holding affects the whole, and whether a shared event could hurt several of the largest positions together.
Why a target mix drifts—and what restoring it costs
In the portfolio rebalancing game, start with 500 tokens in each fictional asset. Reveal a market round and inspect how unequal returns change the weights. A 20% gain in A with no change in B leaves 600 and 500, so A is about 54.55% of the total. The allocation changed even though you did not choose a new target.
Restoring the target reduces that drift but incurs the game's disclosed charge on the pre-fee one-way transfer. Compare with the untouched portfolio under the same shocks. Rebalancing may help on a reversing path or trail when one asset keeps outperforming. It controls exposure; it does not create a guaranteed return premium. Real taxes, spreads, and contribution flows introduce further choices omitted by this short game.
Diversification is not a substitute for sizing risk
A portfolio can be spread across holdings yet still expose essential resources to losses it cannot tolerate. The gambler's ruin experiment explores how repeated bets and stakes interact with a failure threshold. It is not an investment-market model, but it makes an important boundary tangible: surviving long enough to use future opportunities depends on more than an attractive average outcome.
Likewise, borrowed exposure can introduce maintenance demands regardless of how many holdings you own. The margin-call game keeps debt visible while assets move. Diversification and leverage answer different questions. A collection of correlated positions financed with debt can combine a shared loss with forced action, making the financing arrangement part of the risk analysis.
A practical diversification review
Create a one-page exposure map before searching for another holding to add. Record the major weights, common drivers, expected cash needs, and the reasons behind the target mix. Then ask which new information would change that map. If a new holding shares the dominant driver, adding its name may add complexity without addressing the concentration you meant to reduce.
Keep the distinction between model output and evidence. A calculator can show precisely what a chosen correlation implies without showing whether that correlation will persist. Document estimates as estimates, and review exposure when circumstances change. Good learning makes the dependencies legible; it does not supply an allocation that remains appropriate for every reader and every market.
- Which positions dominate the total weight?
- Which underlying events affect several holdings at once?
- How were variability and correlation estimated?
- Has the allocation drifted from its intended purpose?
- What liquidity, debt, tax, or trading-cost constraints affect changes?
Try the ideas for yourself.
These are teaching models. Follow the assumptions in each experiment; the results are not real-world forecasts.
Sources & further reading
Current-event context was checked on October 14, 2026. Follow the original source for newer updates. Worked scenarios are illustrative unless explicitly identified as reported data.
- SEC: asset allocation and diversification ↗
Official investor education; individual suitability is outside this tutorial.
- Markowitz: Foundations of Portfolio Theory ↗
Primary lecture on portfolio theory; Modic's two-asset board isolates one covariance relationship.