Data analysis
Two investments show a low correlation in your backtest. That tells you how their returns moved during the period you measured. It does not tell you how much you could lose in a bad month. For that, you need to examine the mix, its shared drivers and its losses.
The measure
Correlation between two return series ranges from minus 1 to plus 1. Near plus 1, they often move together. Near minus 1, they often move in opposite directions. Near zero means little linear relationship is visible. It does not mean the investments are independent.
The number belongs to a period and a sampling frequency. Daily returns may give a different result from monthly returns. Currency and the treatment of dividends also matter. A chart that reports only “0.2” hides most of the conditions behind the estimate.
Correlation describes past co-movement. It is not a fixed trait of two investments.
The portfolio
If two holdings do not always swing together, their combination can reduce variation in the whole portfolio. Their weights and their own volatility also matter. A small stable holding has a different effect from a large volatile one, even at the same correlation.
Diversification belongs to the whole mix. Correlation is an input to the risk calculation, not the finished answer.
Stress
Take a 50:50 portfolio. Holding A falls 20% in a crisis month and holding B falls 15%. Before costs and rebalancing, the portfolio loses 17.5%. A low correlation over previous years cannot rule out that shared bad month.
There is relevant evidence, but it needs a narrow reading. Longin and Solnik found higher correlation among international equity markets in bear markets. Their study covers those markets and that relationship. It does not assign one crisis correlation to every possible mix of stocks, bonds or property funds.
Analysis
I would start with correlation over the full period. Then I would calculate rolling correlation, for example using 36 monthly returns at a time. That shows how the relationship changes. Finally, inspect the months when the portfolio lost the most: what did each holding do then?
The window is not a magic setting. A short one reacts quickly but moves around more. A long one smooths the series and can hide a recent change. Match dates, currency and the type of return. Dropping missing months without checking them can distort the result.
Costs and rebalancing belong in a portfolio simulation. Correlation alone covers neither.
Shared drivers
Two ETFs do not necessarily give you two different sets of risks. Look at their holdings, sectors, regions and economic drivers. Funds with different names may own many of the same shares. Both equities and property funds can be hurt by higher financing costs.
The weights matter too. A portfolio with 90% in one holding and 10% in another still depends heavily on the first, even if their return series had little correlation in the past.
Ask which event could hurt both positions at once and how many euros the whole portfolio would lose.
Backtesting
The asset-class backtester article compares return and risk for individual asset classes. The version described there does not yet calculate the portfolio effect of combining them. Two separate lines therefore do not establish a tested diversification benefit.
To test a mix, you need returns aligned by date, fixed or documented changing weights and the combined portfolio return series. Then you can compare volatility, maximum drawdown, rolling correlation and the worst months.
Before interpreting the result, check data quality. Different start dates or missing distributions can create an apparent advantage.
The comparison provides the inputs. A claim about a mix also needs the combined portfolio return series.
Read next
Sources