In the past year I had the opportunity to delve back into portfolio construction for some clients, and after three years away from managing portfolios I found it genuinely challenging to understand the new rules of the engineering, which had shifted further than I expected in what felt like a short absence. To be a good allocator you need to hold a vision about the world, about where it is going and about what actually drives markets, and when I decided to go on the investing world tour I thought that visiting all of those places would help me understand the world better and build a vision that sat closer to the truth. In practice the opposite happened, because the more information I collected the more confused I became, and holding a vision steady in a world that keeps pulling plot twists turned out to be considerably harder than forming one in the first place.

What made it harder still is that the market as a whole, and I am not talking about single stocks here, increasingly reacts to something that has become decorrelated from the reality on the ground, so that the evidence you gather by actually going somewhere and looking has a weaker and weaker claim on the price. Almost everything I learned at the beginning of my career about how portfolios are supposed to behave no longer seems to hold in the way it was taught, and I have come to believe that investing today, much like the technology sector it is currently obsessed with, requires you to evolve and adapt rather than to defend the framework you inherited. So let us try to get better at this together by working through a series of questions, and by checking each of the received rules against fifty years of data rather than against memory.

The five rules I want to test are the ones every allocator I speak to can recite without hesitation, namely that bonds hedge equities, that gold hedges crisis, that commodities hedge inflation, that international equities hedge home bias, and that a slice of private assets will lower the correlation of the whole. Almost all of it, as it turns out, was measured inside a twenty three year window that has since closed, and the rules were never wrong so much as they were conditional, with nobody having bothered to write the conditions down.

The one thing to understand before the tables

Correlation is not a property of an asset, it is a property of the macroeconomic regime that the asset happens to be sitting in, and once you accept that single reframing almost everything else in this piece follows without further argument.

The variable that governs the whole structure is the relationship between inflation and the output gap, and when the two move in opposite directions you are living in a demand shock world where a recession brings disinflation, the central bank cuts, bonds rally while equities fall, and diversification arrives free of charge, which is a fair description of the period between 1998 and 2021. When inflation and the output gap move together instead, you are living in a supply shock world where the recession arrives with inflation attached, the central bank cannot cut even if it wants to, and every asset being discounted by the same rate falls at the same time, which describes the 1970s and describes the present.

Campbell and his co-authors documented that the correlation between inflation and the output gap was negative from roughly 1979 to 2001 and then turned positive, and that the stock and bond correlation changed sign alongside it, so the mechanism is neither hidden nor especially controversial once you look at it directly. What I find genuinely surprising is how few allocation frameworks encode it anywhere, given that it determines whether the largest single position in most balanced portfolios is doing its job or quietly failing.

Rule one: do bonds still hedge equities?

This is the foundational assumption underneath everything else, and it is also the one that has failed in the most visible and most expensive way.

Source: Financial Analysts Journal (2024), Investment Research Partners, Future Standard

It is worth reading that table twice, because the negative correlation regime that made the balanced portfolio famous lasted twenty three years and was the longest such period ever recorded in any country, which means that anyone who began a career after 1998 spent their entire professional life inside the exception and drew the entirely reasonable conclusion that they were looking at the rule.

The cost of getting this wrong is substantial rather than marginal, since academic work indicates that a shift in the correlation from minus 0.5 to plus 0.5 raises the volatility of a balanced portfolio by close to twenty per cent and its maximum drawdown by roughly thirty per cent, and the realised evidence matched that model almost exactly when in 2022 the S&P 500 returned minus 18.1 per cent, the Bloomberg US Aggregate fell 13.0 per cent in the worst year in the index’s history, and a standard sixty forty allocation lost approximately 16.7 per cent.

There is a detail here that tends to get left out of the more alarmist versions of this argument, which is that Vanguard’s work suggests a sixty forty investor would only need to move to about sixty two per cent equities in order to reach a comparable risk and return outcome inside a positive correlation regime, so the portfolio is not broken in any structural sense. The difficulty is that the fix requires actively recognising which regime you are in and acting on that recognition, which is precisely the thing a static allocation is designed not to do.

A serious counter argument also deserves stating, because LPL and others hold that bonds were never negatively correlated with equities as such and are instead correlated with the business cycle and specifically with growth shocks, on which reading 2022 was a stress test rather than a structural break and the drivers that established the negative regime remain broadly intact. Both readings fit the data reasonably well, and the reconciliation is straightforward once you separate the two cases, because bonds hedge growth shocks and fail to hedge inflation shocks, which means the question facing an allocator is not whether bonds diversify at all but rather which kind of shock they expect to face and whether they own a second diversifier capable of handling the other kind.

Rule two: does gold still hedge everything?

Gold is the asset where the conclusion most people hold turns out to be roughly right while the reasoning underneath it has quietly become wrong, which is a more interesting position to be in than simply being mistaken.

The crisis hedge does hold, and the record supports it without much qualification, since gold rose 5.5 per cent through 2008 while the S&P 500 fell 38.5 per cent, rallied again in 2020 while equities panicked, and produced close to perfect negative correlation through the stagflation of the 1970s. Over long horizons the equity and gold correlation sits near zero and has drifted into the 0.4 to 0.5 range against non US equity indices as global assets have become more broadly correlated with one another, which is still a materially better diversification profile than most of the alternatives examined in this piece.

The inflation hedge, by contrast, is considerably weaker than most investors assume, because work extending back to the early twentieth century finds that broad commodities beat gold on average during inflationary episodes and that gold was only really decisive during the 1970s, and because more recently gold rose steadily through the disinflation of 2023 to 2025 while market implied inflation expectations remained broadly stable. Then came January 2026, which settled the argument in a manner that no backtest could have managed on its own.

Gold reached approximately $5,595 on 29 January and then fell more than ten per cent within thirty hours, while silver, which had touched an intraday record above $121, fell approximately 31 per cent in a single session on 31 January in its steepest one day decline since 1980, and gold finished the first half of the year down while realised inflation was rising on the back of an energy shock. An asset that falls during an inflation surge is, on that occasion at least, doing something other than hedging inflation, whatever the marketing material says.

The trigger for all of this was not fundamental in any meaningful sense, since it combined the Warsh nomination repricing rate expectations, margin increases at both CME and Shanghai, month end futures dynamics, and index rebalancing that mechanically forced selling as gold’s weight fell back through its cap inside the Bloomberg Commodity Index.

What has actually changed is the identity of the marginal buyer, and this is the part that matters for anyone sizing a position today. Gold pays no yield and therefore historically carried an inverse correlation to US real yields, yet after March 2022 that relationship stopped working entirely, with real yields rising and gold rising anyway, because central banks across Asia, the Middle East and Eastern Europe have been accumulating at a pace not seen in decades, with net official buying up 132 per cent between 2021 and 2024. Those buyers are structurally indifferent to Treasury yields since they are acting on reserve policy and dollar exposure rather than on opportunity cost, which means gold has stopped being a real yield instrument and has become a sovereign risk and currency debasement instrument whose marginal price is set by reserve managers, a different exposure with different failure modes that explains both why gold has beaten Treasuries every year since 2023 and why it behaves so erratically around inflation prints.

One practical note belongs here, which is that silver is not simply gold with more upside, since its industrial demand component and much thinner market make it a leveraged expression of the same underlying trade, as January demonstrated at considerable cost, and it therefore has no business being described as portfolio insurance in any client document.

Rule three: do commodities still hedge inflation?

This is the textbook relationship that survived the fifty year test in the best condition of any examined here, and it deserves rather more attention than it currently receives in European private portfolios.

Goldman Sachs Research examined five separate inflationary periods over the past fifty years, covering the early 1970s oil embargo, the Iranian Revolution, China’s boom in 2005, the late cycle boom of 2007 to 2008, and the post pandemic recovery from 2021, and found that commodities outperformed both equities and bonds in every one of them, with the result holding when the test is restricted to inflation surprises above one percentage point. Commodities also carry low correlation to both equities and bonds while carrying a higher correlation to inflation than either, and the negative correlation to bonds is the genuinely valuable part of that profile because it means the exposure performs in precisely the scenario that broke the balanced portfolio in 2022.

The first half of 2026 handed us an unusually clean natural experiment, since joint US and Israeli strikes on Iran began on 28 February, Iranian forces declared the Strait of Hormuz closed on 4 March with vessel traffic collapsing by around seventy per cent, and Brent moved from roughly $72 in late February to test $100 within days before trading near $120 at the peak.

Source: Bloomberg Professional Services, midyear commodity review 2026

The complex behaved more or less exactly as theory predicts it should, though what theory does not prepare you for is the dispersion inside it, since precious metals led the field in 2025 and finished at the bottom of this table while energy led by nearly forty points, which means that treating commodities as a single undifferentiated exposure would have missed a spread of more than forty percentage points inside one index.

Three caveats attach to all of this and none of them should be skipped. The relationship holds for broad futures indices rather than for single commodities, the inflation exposure is close to linear so that commodities lose on disinflation surprises just as reliably as they win on inflation surprises, with Neville and co-authors finding that in deflationary regimes representing eighty one per cent of the sample since 1946 the real returns annualised to just one per cent, and the volatility of the asset class is equity like with lost decades that are far from uncommon.

Rule four: do international equities still diversify?

Source: Cambridge Associates, CFA Institute, T. Rowe Price

The deterioration recorded in that table was structural rather than cyclical, because as supply chains integrated and revenue bases converged across borders the underlying cash flows being discounted became increasingly common to all markets, with rolling ten and twenty year correlations eventually settling somewhere between 0.80 and 0.90 and staying there.

There is credible evidence that the trend has now turned, since T. Rowe Price found that across 2022, 2023 and 2024 the average correlation between nearly every major index declined relative to the 2015 to 2020 period and attributed the change to deglobalisation and market fragmentation, while emerging markets have consistently offered more diversification than developed ones, with Morningstar showing Emerging Europe falling from a three year correlation of 0.82 at the end of 2021 to 0.16 at the end of 2023.

So far this reads as encouraging news for anyone building a geographically spread portfolio, and the first half of 2026 is where it stops reading that way. South Korea gained approximately thirty per cent over the period, Peru twenty six, Brazil twenty one and Turkey twenty one, against 9.6 per cent for the S&P 500, while emerging markets recorded thirteen positive months out of fourteen and nine consecutive positive weeks in a streak last seen in 2005, and none of that was diversification in any sense that would survive scrutiny.

The Philadelphia Semiconductor Index rose approximately ninety per cent over the same period, roughly twelve times the gain of the S&P 500, while South Korea and Taiwan rose approximately 110 per cent and sixty per cent respectively, which means that an investor who bought emerging markets for geographic diversification in 2026 was in reality buying a leveraged position in the same artificial intelligence capital expenditure cycle that already dominated their domestic allocation. Geographic labels have therefore become poor proxies for factor exposure, MSCI EM has effectively become a leveraged proxy on capital expenditure in one industry, and any diversification analysis conducted at country or region level will systematically understate the true concentration sitting in the portfolio.

Rule Five: is bitcoin still digital gold?

This one was marketed to institutions on an explicit correlation claim rather than on a return claim, which makes it unusually easy to falsify, and the claim is now empirically dead with a date of death that can be established with reasonable precision.

Source: CME Group, LSEG, Phemex, Reuters

The cause of that convergence is structural rather than sentimental, since more than $56 billion in cumulative ETF net inflows moved a significant share of bitcoin’s marginal price setting into regulated institutional products, with BlackRock’s IBIT alone crossing $90 billion in assets by October 2025, so that bitcoin is now priced, hedged and liquidated inside the same operational framework as equity ETFs, by the same portfolio managers, sitting inside the same risk budget. When those risk budgets shrink the asset is sold alongside Nasdaq futures, which happens not because the two instruments resemble each other in any economic sense but because they occupy the same allocation bucket at the same desks, and that is a permanent change in market microstructure rather than a temporary alignment of sentiment.

The first half of 2026 confirmed all of this in the open, since crypto assets fell approximately thirty six per cent while gold held its structural bid, bitcoin declined roughly fifty per cent from its October 2025 record near $126,200 while tracking rate fears and artificial intelligence valuation concerns in lockstep with the Nasdaq, and when the Kospi fell ten per cent overnight on a memory chip unwind bitcoin duly went with it.

One nuance is worth preserving from that same period, which is that crypto equities, meaning the miners, exchanges and infrastructure providers, returned approximately twenty three per cent over the same half in which crypto assets fell thirty six per cent, representing the widest split of the current cycle and demonstrating that the token trade and the equity trade have genuinely separated and should stop being treated as a single allocation decision. Bitcoin may therefore still merit a position on a return basis for investors who want it, but it does not merit one on a diversification basis, and anyone currently holding it as insurance against an equity drawdown is holding the wrong instrument for that particular job.

The finding that actually matters

Everything above operates at the asset class level, and yet the most significant correlation development of the past three years sits one level below that, where most portfolios have not priced it at all, because a single industry has quietly become a macroeconomic variable.

The numbers make the case without needing much help from me, since the Philadelphia Semiconductor Index rose approximately ninety per cent in the first half of 2026 against 7.5 per cent for the S&P 500, SanDisk rose 858 per cent over twelve months, Intel 278 per cent and Astera Labs 434 per cent, while Samsung, Micron and SK Hynix became the tenth, thirteenth and fourteenth most valuable public companies in the world and passed both Berkshire Hathaway and JPMorgan on the way. The previous leadership inverted over the same period, with Microsoft falling twenty per cent in June alone in its worst month since 2008 and Oracle falling thirty per cent, while software judged vulnerable to substitution by artificial intelligence fell considerably harder, taking Intuit down approximately fifty per cent, CoStar fifty seven and The Trade Desk fifty two.

One factor now drives the American technology sector, the emerging market equity complex, industrial metals demand and a meaningful share of global electricity demand growth simultaneously, and that factor is the durability of capital expenditure on artificial intelligence, which was projected to reach approximately $805 billion in 2026 against $449 billion in 2025 while consuming roughly ninety three per cent of hyperscaler operating cash flows compared with thirty three per cent as recently as 2023. Consider then an allocator holding US large cap equity, emerging market equity, an industrial metals sleeve and an infrastructure or data centre allocation, which reads as four diversified exposures on the fact sheet and functions in the current regime as four expressions of a single bet, and note that June 2026 demonstrated this in public when semiconductors, emerging markets and crypto all fell together while bonds and REITs held.

The illiquidity illusion

Private assets show low correlation to public markets in essentially every capital market assumptions document ever published, and that figure should be treated as a measurement artefact until somebody proves otherwise.

Private funds update their valuations quarterly or around capital events rather than continuously, with managers relying on internal models, comparable pricing or external appraisal, all of which diffuse sharp price movements across multiple reporting periods and leave a statistical footprint in the form of serial correlation, where current reported returns depend heavily on prior reported returns. The consequence is that smoothed returns systematically understate volatility, inflate Sharpe ratios and reduce the apparent correlation to public markets, and Verus states the problem candidly in its own methodology when it notes that private equity and private real estate correlations are especially difficult to model owing to appraisal based pricing and lag, requiring factor model estimates for the former and de lagging of quarterly returns for the latter.

None of this is new, given that Geltner published on appraisal smoothing in 1991 and that Getmansky, Lo and Makarov published on serial correlation in 2004, and what has changed is that it has become a fiduciary question rather than a technical footnote, with the August 2025 US executive order expanding access to alternatives inside 401(k) plans and the Department of Labor’s proposed rule of 31 March 2026 on selecting designated investment alternatives both raising the measurement problem directly.

I want to be fair to the asset class here, because the honest version of this argument is considerably more interesting than the dismissive one, and private assets do carry structural features that public equivalents genuinely lack, including illiquidity premia, income driven return profiles and leverage applied at the asset rather than the portfolio level. They also provide real behavioural diversification, since investors cannot panic sell what they cannot price daily and that restraint is worth actual money over a full cycle, while the dispersion within the category is genuine too, with listed REITs returning approximately 9.75 per cent in the first half of 2026 while private real estate as measured by NFI-ODCE lagged materially behind. All of that is defensible on its own terms, and what is not defensible is arguing for the allocation on the strength of a correlation coefficient that the valuation methodology has largely manufactured.

What still works

Having spent most of this piece taking the received framework apart, it is only useful if I am equally specific about what actually survived both 2022 and 2026, and the pattern is that the exposures which behaved differently from equities when it mattered were defined by what they do rather than by what they own.

In the first half of 2026 managed futures returned 9.28 per cent and hedge funds broadly returned 9.22 per cent, while in the high oil regime specifically managed futures returned 9.1 per cent and global macro commodity strategies returned 8.8 per cent, and the dispersion inside the category was extreme in a way that favoured specialists over scale. Smaller specialist funds comfortably beat the largest multi strategy platforms over that period, with CastleKnight’s event driven fund returning 42.3 per cent, Whale Rock 72.5 per cent, TAL China Focus 95.1 per cent and Keystone 62.7 per cent, while discretionary global macro delivered 17.9 per cent across 2025 in its strongest year since 2009.

Three principles follow from that table and from everything preceding it. The first is that you should diversify by shock type rather than by asset label, owning at least one exposure that works against growth shocks and one that works against inflation shocks, since between 1998 and 2021 a single instrument was able to perform both jobs and it no longer can. The second is that concentration must be measured at the factor level, because country, sector and asset class labels have all become unreliable proxies and the relevant questions in 2026 concern exposure to capital expenditure on artificial intelligence, to the energy supply chain and to the discount rate, rather than the split between domestic and international equity. The third is that correlation inputs should be treated as regime conditional, since a covariance matrix estimated across a ten year window that spans two opposite regimes describes neither of them accurately, and where possible you should estimate separately by regime and state plainly which regime the allocation assumes.

What this looks like as a portfolio

Frameworks are easy to agree with and considerably harder to act on, so here is the shape that the evidence points towards, offered as an illustration of the logic rather than as a recommendation to anybody. It is euro based, balanced in risk profile, built for a five year horizon, and every weight below is a starting point that a real mandate would move in one direction or another.

Three things distinguish this from the standard sixty forty plus alternatives model that the large houses are currently publishing, and each of them follows directly from the evidence set out above.

The inflation sleeve sits at fifteen per cent rather than at zero, which is the entire argument of this piece expressed as a weight, and it is also the single most common structural gap I encounter in European private portfolios, given that J.P. Morgan’s illustrative sixty forty plus allocates 7.5 per cent to real assets inside a much larger alternatives block that is dominated by private markets and therefore doing a rather different job.

Equity sits at forty five per cent rather than at sixty, with total portfolio risk remaining broadly comparable because the ballast has been spread across three shock types rather than concentrated in one, and what changes is the failure mode rather than the volatility, since a conventional sixty forty has only one way to break and has now broken twice in four years.

Emerging markets are held at seven per cent, which on valuation grounds alone is too low and which I would happily raise under different conditions, but until the semiconductor factor exposure normalises the addition of emerging market equity mostly adds the same bet that is already sitting inside the developed sleeve, as the first half of 2026 demonstrated to anyone paying attention.

Three obvious adjustments follow from there, in that higher spending requirements would move five per cent from commodities into infrastructure and short duration, genuine tolerance for illiquidity would take the private sleeve to twelve per cent funded from developed equity, and a hard recession or deflation view would take duration to twenty two per cent while cutting commodities to four.

On gold there is one constraint worth respecting even if you disagree with the house that published it, since J.P. Morgan’s own guidance holds that gold’s volatility is closer to that of an equity than a bond and caps the position at mid single digits inside a balanced portfolio, which is a useful counterweight to the maximalist gold case currently doing the rounds and is the reason the sleeve above sits at five rather than at ten.

What I take from this

The period from 1998 to 2021 was anomalous in almost every respect examined here, since stocks and bonds moved inversely for the longest stretch on record, inflation remained low and counter cyclical throughout, globalisation was simultaneously raising cross market correlation and suppressing volatility, and the credibility of central banks was assumed rather than seriously tested at any point. Every diversification assumption formed inside that window was calibrated to conditions that no longer hold, and the window was long enough that an entire generation of allocators, including me, formed their working instincts inside it without ever being told that the conditions were temporary.

The current environment resembles the 1970s far more closely than it resembles the 2010s in its underlying mechanics, with supply shocks dominant, inflation running pro cyclically, geopolitics repricing energy on a monthly basis and fiscal expansion constraining whatever the monetary response might otherwise have been. In that environment the textbook itself says that commodities should work, that bonds will not reliably hedge equities and that equity valuations should de rate, and the evidence from both 2022 and 2026 is consistent with all three of those predictions.

The constructive part, and the reason I have found this exercise worth the time rather than merely dispiriting, is that this is a knowable problem rather than an unknowable one, since the regimes are identifiable in real time from the relationship between inflation and the output gap and the assets that work inside each of them are thoroughly documented. What is actually required is that the allocation be permitted to change when the regime does, which is a governance question at least as much as an analytical one and which, in most institutions I have worked with, turns out to be the harder of the two by a considerable margin.

The most expensive assumption sitting in portfolios today is therefore not that any particular correlation will continue to hold, but rather that the covariance matrix is a stable input at all, and almost everything else in this piece follows from that single inherited error.

Influidence is an independent investment research firm based in Porto. We combine field based intelligence with quantitative analysis for asset managers, family offices and allocators.

This article is for informational and research purposes only. It does not constitute investment advice, a recommendation, or an offer to buy or sell any security. The illustrative allocation shown is a generic framework produced to demonstrate the analytical logic of the piece. It takes no account of any reader’s objectives, financial situation, tax position, liabilities or liquidity requirements, and it should not be implemented without independent professional advice. Past performance is not a reliable indicator of future results, and the correlation relationships described here are historical observations that may not persist.

Sources

Stock and bond correlation: Financial Analysts Journal, “Empirical Evidence on the Stock–Bond Correlation” (2024); Vanguard, “Understanding the dynamics of stock/bond correlations”; LPL Research (December 2025); Investment Research Partners (July 2026); Future Standard.

**Gold and precious metals:** LSEG and FTSE Russell, “Gold in a fragmented world”; CAIA, “Inflation Hedging in Strategic Asset Allocations”; FXStreet (June 2026); VanEck gold outlook; Rothschild & Co (February 2026).

Commodities: Goldman Sachs Research, “Which commodities are the best hedge for inflation?”; Bloomberg Professional Services midyear commodity review 2026; Neville, Draaisma, Funnell, Harvey and Van Hemert (2021); IEA Oil Market Report March 2026.

Geographic diversification: Cambridge Associates; CFA Institute; T. Rowe Price (October 2025); Morningstar; Allianz Research, “The semiconductor premium” (June 2026).

Digital assets: CME Group rolling correlation research; Phemex (April 2026); Bitwise Quarterly Crypto Market Review H1 2026.

Private markets: Geltner (1991); Getmansky, Lo and Makarov (2004); Verus 2026 Capital Market Assumptions; US Department of Labor proposed rule, 91 Fed. Reg. 16088 (31 March 2026).

Alternatives: PivotalPath regime analysis; Hedgeweek; RCM Alternatives Asset Class Scoreboard June 2026.

Allocation frameworks referenced: J.P. Morgan Asset Management, 2026 Long-Term Capital Market Assumptions and Global Asset Allocation Views 1Q 2026; BlackRock, “Rebuilding 60/40 portfolios with alternatives”; LPL Research, 2026 Strategic Asset Allocation.