Rational
Signals

The ETF industry will tell you that more choice is good for investors. More products, more precision, more ways to express your investment view. What they won't tell you is that most of that choice is an illusion — and a profitable one, for them. There are now over 9,000 ETFs available globally, managing over $20 trillion in assets. Every week, new funds launch with names that sound distinct — "next generation," "innovation," "thematic," "smart beta," "ESG-tilted." The marketing is sophisticated. The underlying holdings are frequently identical.

One of the most common problems is investors holding too many funds that own the same underlying stocks — a global equity ETF, an S&P 500 ETF, and a technology ETF that together hold Apple, Microsoft, and Nvidia three times over. On paper the portfolio looks diversified. In practice it is the same bet, repackaged and relabelled. When you own five ETFs, what do you actually own? Most investors cannot answer that question cleanly. You see five lines on a statement, not the five hundred stocks behind them. You cannot easily see your true exposure to any single company, sector, or geography across your whole portfolio. You cannot see the overlaps. You cannot see the concentrations building quietly underneath. You can't manage what you don't measure — and right now, most investors are not measuring the right things.

At Rational Invest, we believe every investor deserves a clear, complete picture of what they actually own — not just the funds that hold it. Before adding any ETF to your portfolio, look at what it actually holds, not what its name suggests it holds. If the top ten holdings look familiar, they are. You are not buying diversification. You are buying a story. Sector and thematic ETFs have their place — there are good reasons to make a specific bet on energy, healthcare, or artificial intelligence. The problem is volume. At some point, the new fund serves the asset manager far more than it serves you. The ETF industry has made investing accessible to everyone. It has also made it very easy to own the same thing many times over — and pay for it every time.

For thirty years, the smartest line about financial power belonged to a political consultant. "I used to want to come back as the President or the Pope," James Carville said in the 1990s. "But now I'd come back as the bond market. You can intimidate everybody." He was right then. He is more right now. Government borrowing costs have surged across the US, Japan, the UK and Germany — not because of any single policy decision, but because investors are worried on two fronts: that inflation may stay higher for longer, and that governments are borrowing too much with no credible plan to stop.

The distinction that matters: central banks control short-term rates. But longer-term borrowing costs are set by millions of investors every day, asking one simple question: are we being paid enough to lend to this government for this long? When the answer becomes no, no politician can reverse it by making a speech. Only by changing the underlying reality.

To understand why, you need to go back to 2008. The global financial crisis triggered fifteen years of near-zero interest rates and quantitative easing — central banks pumping money into the financial system to prevent economic collapse. The pandemic in 2020 accelerated this further, with governments and central banks unleashing the largest peacetime stimulus in history. The bill arrived in 2021 and 2022, when all of that money collided with supply chain disruptions and surging demand, pushing inflation to multi-decade highs. Central banks had no choice but to respond — pivoting abruptly from emergency stimulus to the fastest and most synchronised cycle of rate hikes in forty years. Fifteen years of accumulated assumptions about cheap money and low rates were unwound in eighteen months.

When inflation surged after the pandemic and central banks raised rates at the fastest pace in forty years, long-term US government bonds lost more than 30% in a single year — the iShares 20+ Year Treasury Bond ETF fell 31% in 2022 alone. Stock markets fell at the same time, destroying the central promise of the classic balanced portfolio — that bonds would cushion the blow when stocks fell. The Vanguard LifeStrategy Moderate Growth Fund, a typical 60% shares and 40% bonds portfolio, lost 16% that year. Investors holding long-term government bonds weren't running a conservative position. They had locked in near-zero returns when rates were at historic lows — and when rates rose, the market value of those bonds collapsed. They earned almost nothing, and lost a great deal.

The conditions that made bonds reliably safe for four decades — falling inflation, falling rates, and central banks pumping money into the financial system — are not obviously returning. Deficits remain large and politically difficult to reduce. The industry's justification was always "diversification" and "capital preservation." But the asymmetry was stark: fund managers were collecting full fees to put client money into bonds paying 1% or less, with enormous losses waiting if rates ever rose. In 2022, clients found out. Governments don't control long-term borrowing costs. Markets do. And markets are doing what they always do when confidence erodes: making it more expensive to be reckless — slowly at first, then very quickly indeed.

DeepSeek is one of the most unusual AI success stories of the decade. It did not emerge from a Silicon Valley tech giant or a leading university, but from High-Flyer, a Chinese quant hedge fund founded by mathematician and trader Liang Wenfeng. Long before the current AI boom, High-Flyer was already using machine learning for trading and research. In 2023, the fund spun out DeepSeek as an independent AI lab. The result surprised Silicon Valley: its models appear to rival leading Western systems, despite US chip restrictions. Claims that it achieved this with far fewer resources are, however, overstated — High-Flyer had accumulated significant stocks of advanced Nvidia chips before US export controls tightened in 2022.

DeepSeek nonetheless reflects a broader shift in China's AI ecosystem. Constraints on access to cutting-edge US hardware have forced firms to prioritise efficiency, while Beijing's push for technological self-reliance has accelerated the development of domestic alternatives. That shift is becoming visible in technical choices: DeepSeek has begun optimising its latest models for Huawei's chips, part of China's effort to build a more self-contained AI stack. Ironically, restrictions designed to slow progress may be accelerating a parallel wave of innovation.

The success of DeepSeek, born inside a quant hedge fund, suggests that finance is no longer merely allocating capital to technology — in some cases, it is becoming the laboratory itself. It is also a working demonstration of something we believe deeply at Rational Invest: that competitive advantage in this decade will not come from doing more, but from building systems that know precisely when to act, and why. DeepSeek didn't win by throwing resources at the problem. It won by engineering out the waste. That is exactly the discipline we are applying to portfolio intelligence.

It is 2035. You no longer log into your finances. There is no banking app to check, no investment account to open, and no financial advisor to call. Instead, there is simply a conversation. "Can I afford to spend six months a year in Italy?" "What happens if I retire at 58?" "Can I buy the boat without regretting it?" Your AI already knows your salary, mortgage, spending habits, taxes, savings, investments, insurance, and long-term goals. It knows how you react when markets fall, how much risk you are truly comfortable with, and probably that you always underestimate the cost of holidays.

That future suddenly feels much closer after OpenAI this week launched new personal finance tools inside ChatGPT, currently available only to users in the United States — allowing it to connect to financial accounts, analyse spending and investments, and answer personal finance questions. For the first time, a major AI platform is trying to become the main connection between consumers and the financial system itself.

In 2035, your AI may automatically lower your taxes where possible, negotiate mortgage rates with banks in real time, and continuously adjust your investments based on your goals and behaviour. Of course, this raises difficult questions. Will people trust AI with their life savings? Will regulators allow AI systems to move money automatically? Who becomes responsible when an AI gives bad financial advice? But history suggests convenience usually wins. People once resisted online banking, then mobile banking, then digital wallets. Eventually, things that once felt risky became normal. And if AI finance follows the same path, the biggest transformation may be that finance itself quietly disappears into the background of life — becoming less like something people actively manage, and more like electricity: always there, intelligently optimised, and mostly unseen.

Trump's arrival in Beijing this week — the first US president on Chinese soil in nearly a decade — is the kind of event that looks like diplomacy but functions as a signal. A declining dollar reserve currency power, debt-to-GDP north of 125%, and a geopolitical transition already underway, all pointing to the same portfolio conclusion — that a US-centric 60/40 allocation is a bet on American monetary hegemony persisting indefinitely, which history flatly does not support.

When the sitting US president flies to the Great Hall of the People to negotiate trade terms, rare earth access, and Iran policy from a position of visible need, that's not a diplomatic visit — it's a data point. The question for investors isn't whether the transition is happening; it's whether your asset allocation has caught up with the geography.

Our base case is a multipolar transition: US dominance eroding at the margin, Chinese power expanding but constrained by structural headwinds. The investment implication is not a rotation from West to East — it is a diversification across poles.

In long-term multi-asset allocation, currency exposure is the risk that most investors price last and feel first. 2025 provided a precise illustration. The S&P 500 delivered a total return of 17.9% in US dollar terms — a strong year by any historical standard. Yet a European investor holding the same index without currency hedging earned approximately 4% in euro terms, after the dollar weakened 13% against the euro over the same twelve months. Same index. Same companies. Same earnings growth. Roughly one quarter of the return in the currency that actually matters to a European investor.

It is true that over very long horizons — fifteen to twenty years or more — currency fluctuations between major developed market pairs tend to mean revert, and the case for leaving foreign equity unhedged over a full investment cycle is academically defensible. But mean reversion at the portfolio level does not solve the problem at the investor level. Most investors have real liability points along the way — a retirement date, a drawdown period, a property purchase — where the exchange rate is not a theoretical number but an actual one.

The correct anchor for any currency framework is not the base currency of the portfolio, but the currency in which the end client has, or will have in retirement, their primary expenses. A multi-asset portfolio without an explicit liability-currency framework carries hidden concentration risk, because every unhedged foreign asset is simultaneously a bet on that asset and a bet on its currency relative to what the money is ultimately for. Currency risk should be understood, sized, and managed.

In December 1989, the Japanese Nikkei 225 peaked at 38,915. It would not see that level again for thirty-five years. An investor who bought at the top in 1989 waited through the entire careers of an entire generation of fund managers, through the dot-com boom and bust, through the global financial crisis, through Abenomics, through a global pandemic — and finally broke even in 2024. Not a profit. Even. Thirty-five years of patience rewarded with a return to zero in nominal terms, and a deeply negative return once inflation is accounted for.

The Japan story is not a historical curiosity. It is the single most important stress test of every assumption that underpins conventional long-term investing — that markets always recover, that time in the market beats timing the market, that patience is always rewarded. In Japan's case, patience was necessary but not sufficient. What mattered was not how long you held, but what you paid when you bought, what you owned when you bought it, and whether your portfolio was sufficiently diversified across geographies and asset classes to survive a generation-long drawdown in a single market.

This is the core of the Rational Invest philosophy. We do not believe that long-term investing means buying an index, closing your eyes, and waiting. We believe it means building portfolios that are genuinely diversified across asset classes and geographies, priced rationally at entry, and constructed with an explicit understanding of the scenarios in which conventional assumptions fail. Japan in 1989 was not an accident. It was a market priced for perfection — extreme valuations, extreme concentration, extreme consensus — and the thirty-five years that followed were the price of that perfection. The rational investor's job is not to predict when the next Japan will happen. It is to ensure that when it does, their portfolio was never built on the assumption that it couldn't.

The most dangerous financial model is not the one that looks broken — it is the one that looks perfect. AI-generated models have introduced what we call the flipped house problem: a structurally compromised asset dressed in fresh paint. Just as a flipped house is a property purchased cheaply, cosmetically renovated to mask underlying defects — cracked foundations, faulty wiring, compromised plumbing — and sold at a premium to a buyer seduced by surface appeal, AI-generated models present polished formatting, credible logic flows, and professional layouts that create trust before the audit has begun.

Beneath the surface, balance sheets fail to balance, formulas are hardcoded rather than dynamic, and circular references lurk in revolving credit structures waiting to detonate under scenario stress. The deeper risk is not technical — it is cognitive. When a human builds a model, they own every assumption and every link; when an AI builds it, the analyst inherits the illusion of understanding without the understanding itself.

Rational investors should treat AI-generated financial models exactly as they would a flipped house: conduct a full structural inspection before committing capital, and never confuse cosmetic confidence for structural integrity.

If the flipped house problem is the risk, verified data infrastructure is the fix — and the plumbing is finally being installed at scale. Claude now connects natively with LSEG, FactSet, S&P Global, and Morningstar, and has expanded further to include Moody's credit data alongside other data feeds. OpenAI, Google, and others are following suit.

A model anchored to these verified, auditable data stacks — with every output source-attributed and traceable — is a structurally different product from one that has hallucinated its way to a polished spreadsheet. The announcement, made at Anthropic's financial services briefing in New York on May 5th alongside JPMorgan CEO Jamie Dimon, signals that Anthropic is not merely selling AI software to banks — it is building the data infrastructure and deployment layer to become the operating system for institutional finance.

The rational framework has not changed: AI in investment management is only as trustworthy as the data beneath it. What has changed is that the data is now there.

When Anthropic released its ten financial services agents last week, the most revealing detail was not the agents themselves — it was where they published them. Not behind a paywall, not inside a proprietary platform, but on GitHub, open source, under an Apache 2.0 licence, with every system prompt readable in plain English markdown.

The move is not generosity — it is strategy. By publishing the agent instructions publicly, Anthropic solves the single biggest barrier to AI adoption in regulated finance: the black box problem. Compliance officers and risk committees cannot approve what they cannot audit, and every major financial institution deploying AI in production needs to show regulators exactly what instructions the system is running on. GitHub makes that possible.

The customisation layer matters equally — a Goldman Sachs does not want Anthropic's generic pitch agent, it wants one built on its own deal process, its own templates, its own data connectors, and its own approval chains. The open repo makes that straightforward. And with 22,700 stars and 3,100 forks in under two weeks, the ecosystem flywheel is already turning — every firm that forks and customises becomes more deeply embedded in Claude infrastructure, raising switching costs quietly and durably.

The headline said Anthropic launched ten AI agents for Wall Street. The reality is more consequential: Anthropic is making a serious bid to become the operating system that institutional finance runs on.