In June, the Kospi peaked above 9,300 points. It had risen 76% in 2025. It then doubled again in the first half of this year. Retail investors piled in. Leveraged single-stock ETFs on Samsung and SK Hynix launched in late May, letting them magnify their bets.

Then July happened. The Kospi fell 22%. It was the index's worst month since the Global Financial Crisis. Investors who bought the leveraged ETFs near the top and held to mid-July lost about half their money.

It is tempting to read this as a story about Korean retail investors and single-stock leverage. It isn't. It is a story about human behaviour, and it repeats in market after market, time after time.

The pull of the crowd

Herding is not a story about poor judgement. It is a story about a very human need for safety, playing out in markets.

Devenow and Welch's 1996 review of the herding literature sets out two distinct versions of the same behaviour. The first is rational herding, where investors copy others because they assume the crowd has information they don't. Each new buyer adds a little more apparent confirmation, and the move builds on itself rather than on fresh evidence. The second is irrational herding, where investors follow the crowd simply because it feels safer, independent of any information content at all.

Both versions are underpinned by the same need: safety in numbers. Being wrong alongside everyone else feels different to being wrong alone, whether or not there's a good reason for the crowd to be right. The problem is that this feeling of safety and the actual level of risk move in opposite directions. The more crowded the trade becomes, the more comfortable it feels, and the more fragile it actually is.

Two versions of the same pattern

The Kospi shows the sharp version. A fast, leveraged, retail-driven move that unwound just as quickly. FOMO plus leverage plus a crowded trade is a well-worn combination, and it tends to end the same way.

Precious metals show the slower version. Gold and silver's move over 2024 and 2025 was driven by real forces: central bank buying, rate expectations, currency dynamics. But the story around the move was also self-reinforcing. Rising prices attract new buyers, media coverage amplifies the narrative, and at some point the narrative itself becomes a reason to buy, separate from the fundamentals that started it. This is a genuine multi-year trend, not a single blow-off. That is precisely why it is a better test of discipline than a bubble is. Staying in a strong trend is not the mistake. Staying in without any plan for managing the position is.

Where trend following fits

Trend following is a way to trade that behaviour rather than get caught in it. Price sets the entry and the exit. That's what makes it possible to be in a crowded trade without being trapped by it.

Trend following looks for trends early, often before they're conspicuous, and stays with them as the crowd arrives. The Kospi's move and the precious metals run are both trades our systematic models identified before either became a mainstream story.

Herding is fuel. The buying that pushes a trend from early to obvious, and then from obvious to crowded, is exactly what a trend-following strategy is designed to capture.

Systematic trend following sizes positions based on volatility, not conviction. When its indicators show the trend is no longer there, it exits. There's no leverage stacked on top of an already extended move. A system does not need to keep believing in a story to justify staying in.

It also has no memory of its own trades. It doesn't hold on because admitting the trend has turned feels like admitting a mistake. A system has no ego invested in being right, so it has no reason to ignore what the price is doing.

That is the real distinction. Not whether to participate in a trend, but how early the position was built, and how quickly it is closed once the trend ends.

The Kospi and precious metals weren't failures of the crowd, or of the trend. Herding will keep shaping markets. What matters is having a process for participating in it.

Artificial Intelligence remains the dominant force in global equity markets. Investors are captured by the growing capabilities of the most advanced frontier AI models. Meanwhile, the unprecedented scale - and price tag - of the AI infrastructure buildout continues to be a flashpoint, dividing AI optimists and those concerned about whether future returns will ultimately justify this level of investment.

Neither of these headline grabbers can be looked upon in isolation. Instead, investors must understand that these developments share a common objective: to fundamentally transform the way work is conducted across the global economy.

More than words: Judging companies on their results

Early in the AI era, exactly how this revolutionary technology would be deployed and the degree to which it would enhance productivity were largely matters of conjecture. That is no longer the case. The recently concluded earnings season provides quantifiable evidence that AI is delivering on its ambitious promise.

Importantly - and in contrast to recent quarters - management commentary has shifted from AI adoption metrics to AI financial results. Areas in which executive teams highlight AI's financial impact include operating margins, productivity gains, revenue generation, and even physical workflows. And while tech companies, with their emphasis on coding, and industrial concerns, given their headstart in automation, are at the forefront of AI adoption, healthcare, financial, consumer, and materials firms are also reporting how integrating this technology contributes to their bottom line.

This dispersion of AI efficiencies is no surprise to us. We've spoken previously of AI enablers, enhancers, and end users. Given the backlog of AI infrastructure initiatives, the enablers - led by hyperscalers - are still a major part of the story.

But the other two categories have now entered the conversation. This, in our view, has major investment implications, as 1) not all companies will have the same level of success in adopting AI and 2) as was illustrated during the digital revolution a quarter century ago, the greatest share of economic gains was generated by novel technologies' end users. We believe we are on the cusp of a similar transition.

Watching margins

To gauge AI's financial impact on the corporate sector, we deployed - not by coincidence - an AI model to scour the recent earnings reports of 109 listed companies. Of those, 69% disclosed some degree of margin improvement due to productivity gains and AI-associated cost reductions.

While the rate of margin improvements did not surprise us, in many instances the scale did. IBM, for example, stated it realized $4.5 billion in AI-related savings in 2025. As is the case with other companies, IBM has grown more confident in referencing AI in company guidance: the company is targeting these savings to reach $5.5 billion in 2026. Within financials, PayPal is guiding the street toward an expectation of $1.5 billion in run-rate savings, with most driven by AI.

Margin improvement is coming in the form of both operating leverage (Alphabet reports roughly 50% of its internal software code is written by AI agents) and declining product development expense, as illustrated by MercadoLibre reporting that line item has declined by 120 basis points (bps), year over year. These examples highlight the trend that much of the cost savings is occurring within the  tech (coding) and customer service functions. MercadoLibre also announced a 90% resolution rate in AI-based customer service interactions.

Building on its foundation of machine learning, AI is now reaching the factory floor and mining operations. In one of its facilities, Hitachi reduced inventory held by 50% and lead times by 77%. Copper miner Freeport McMoRan reported increasing mill throughput by 10% while reducing unit costs by up to 15%. Within the energy space, Shell announced $400 million in annual savings and reduced unplanned downtime by 45%.

The top line gets involved

It's not just margins that are providing an AI-powered lift to earnings. Within our data set, 26% of companies discussed AI's impact on revenue growth, with most operating digital platforms in the tech and consumer space.

One development highlighted by several companies is digital assistants materially contributing to sales growth. Amazon's assistant, for example, was credited with $12 billion in incremental annual retail sales, and Walmart reported that orders utilizing its shopping assistant were 35% larger than those that did not.

Platforms in which advertising comprises a significant contribution to revenue also indicated that AI improved results. In a telling example, Alphabet's AI Max tool was stated to have increased conversion volume by 15%. On the buyer side, Proctor & Gamble's Auto-Bidder system adjusts ad-bids every 15 minutes in response to real-time retail data. The upshot, according to the company, is a four-fold higher return on its branded sales compared to legacy static bidding.

Working smarter

Many of the workflows likely affected by AI tend to have a large human element. While productivity may be defined as doing more with less, thus far AI is delivering more with the same.

Several companies in our data set reported growth while keeping headcount flat. Commentary reveals that AI has been most effective when being applied to repetitive or high-volume tasks. This is freeing up companies to reallocate labor to more value-added work.

Many companies have slowed incremental hiring, and there have been layoffs, but this is evidence of the emerging trend of growth decoupling from headcount. An example is JPMorgan expecting to reduce consumer banking headcount 10% by 2030 while growing the business 25%. Across sectors, companies reported a shift from limited pilot programs to increasing licenses across their workforce.

While these initiatives are showing up in company financials today, other AI applications are aimed at delivering economic value in the future. Pfizer, for example, stated it expects AI integrated into drug discovery, clinical trials, and manufacturing to increase enterprise value by up to $4 billion over the mid term.

Acting on the evidence

We view these examples as just the beginning. Other AI enhancers and end users will invariably apply the technology in innovative and unique ways. The cadence and breadth of the rollout, in our view, have significant investment implications. Early movers have the potential to gain an advantage over slower-acting peers. Some management teams will underestimate AI's potential for disruption and be left behind. We expect future earnings seasons to shed additional light on this dispersion.

For the first time, investors have the opportunity to analyze real data to gauge AI's financial impact. Given our ethos that equities follow earnings, we believe investors should seek to identify the companies that can translate AI deployment to top-line growth, margin expansion, and ultimately higher earnings.

And it's not only equities markets where AI is a driving - and differentiating - force. Improving margins should be on bond investors' radar as well, as they may improve the coverage ratios of AI-forward companies.

Conversely - and given the magnitude of the AI investment cycle - the hyperscalers that deliver the best value proposition to their customers are likely better positioned to generate the cash flows to pay down debt, while those employing less effective strategies may be left with overleveraged balance sheets.

Commercial real estate debt has moved firmly into the spotlight. Recent reporting on borrower stress, loan defaults, enforcement activity and related-party lending have brought greater scrutiny to lending practices across the sector. While the circumstances differ, the headlines have sharpened investor focus on the risks beneath the broad CRE debt label.

At the same time, the market continues to offer a genuine investment opportunity. Banks have moderated their exposure to commercial real estate, while institutional investors have sought illiquidity premia from floating-rate, asset-backed investments. However, the post-COVID performance of property sectors has diverged materially, including between premium and secondary office assets. In this environment, the quality of the underlying property and the discipline of the lender's underwriting become increasingly important.

That dispersion extends beyond the underlying properties to the managers themselves. Participants range from small private lenders specialising in land and residential construction loans to large institutional managers providing substantial, structured facilities against major commercial assets. Their experience, governance and underwriting capabilities can differ significantly.

For investors, this makes manager selection increasingly important. The relevant question is not whether a manager describes its underwriting process as sophisticated, but whether that process identifies a loan's key vulnerabilities and translates them into appropriate pricing, leverage and structure. This month, that's what we're watching.

Security Is Only the Starting Point

CRE debt has become a growing allocation within Australian private debt portfolios. The appeal is clear: loans are generally secured by physical assets, have relatively short tenors, pay floating-rate interest and often incorporate structural protections tailored to the individual transaction. Together, these features can provide attractive risk-adjusted income and meaningful downside protection. But the protection is only as strong as the underwriting that supports it.

At the heart of private debt investing is a simple question: does the return adequately compensate investors for the risk? Answering it requires a disciplined assessment of the sponsor, asset, market, cash flows and, critically, the collateral value available to support recovery. Underwriting is therefore not merely a step in the credit process, it is the foundation for determining the appropriate price, leverage and structure of the loan.

CRE debt is often described as asset-backed and mortgage-secured, but those labels are no substitute for detailed analysis and judgement. The protection provided by the mortgage depends on the value and liquidity of the underlying property. These, in turn, are shaped by the resilience of the property's income, its physical and locational attributes, market conditions, capitalisation rates and investor demand.

This is where underwriting matters. An office loan can look compelling when the business plan assumes rapid lease-up, stable rents, falling incentives and a favourable exit valuation. But if the underwrite relies on those assumptions being achieved, the lender may have accepted too much leverage for too little return from day one.

Underwriting should not simply test whether the sponsor's strategy can work. It should test how the loan performs when it does not. Put simply: lend on the collateral, not the story. That means applying evidence-based and appropriately conservative stresses to understand where value deteriorates, liquidity recedes and the prospects for recovery become uncertain.

Testing the Assumptions

The starting principle is simple: the underwrite must inform the price, leverage and structure of the loan, not be reverse engineered to support terms already proposed. That requires a qualitative assessment of the sponsor's experience, financial capacity, track record and alignment, together with a clear understanding of the property, its competitive position and the credibility of the business plan.

Quantitative analysis then tests how the loan performs when conditions depart from the business plan. Cash flow modelling should assess the effects of changes in occupancy, rents, incentives, lease-up timing, renewal rates and capital expenditure on the property's income, value and debt position. Effective underwriting combines dynamic modelling, relevant historical market data and an understanding of market cycles to form a coherent view of the loan's resilience and potential recovery. The analysis is rigorous, but applying judgement to its assumptions and conclusions requires a balance of art and science.

Finding the Margin for Error

The real value of underwriting lies in identifying the loan's margin for error. A base case may show that the borrower's strategy can succeed, but not how the loan performs when assumptions move adversely. The analysis should extend beyond a single forecast to test how changes in cash flow, valuation, debt accretion and exit liquidity interact. Four variables are particularly important:

Capitalisation rates

Capitalisation rates are a key driver of property value, particularly where income has not yet stabilised. Even a modest increase can materially reduce value and increase leverage. The underwrite should therefore test exit values under different interest-rate, property-risk and investor-demand assumptions, rather than relying on current capitalisation rates persisting.

Net income

Property income is primarily influenced by occupancy, rents, incentives and operating costs. Rents may remain stable while effective rents decline as incentives or landlord contributions increase. Rising non-recoverable costs, such as insurance premiums and land tax, can adversely impact income and value.

Leasing

Longer lease-up periods delay rental income, increase interest reserve usage and may require additional capital incentives. This can increase debt at the same time as the delayed income reduces collateral value, compounding the effect on leverage. The underwrite should also test whether leases can be renewed on acceptable terms when they expire.

Capital leakage

Make-good costs, leasing fees, tenant works and interest shortfalls consume liquidity before they contribute to property value. Sensitivity analysis should test whether available reserves are sufficient and whether further delays or costs cause interest to capitalise, debt to increase and the lender's equity cushion to erode.

A single downside LVR (Loan to Value Ratio) does not tell the full story. The analysis should show how the loan deteriorates under stress, which assumptions drive that deterioration and how much protection remains when multiple variables move adversely together. Heat maps, breakeven analysis and scenario tables all help identify these pressure points and assist in determining the appropriate leverage, pricing and loan structure.

Turning Expectation into Obligation

Underwriting identifies where a loan may come under pressure. Structuring determines the lender's protections and options when it does. Covenants, review events, cash controls, reserve accounts, reporting requirements, leasing and/or selling milestones and valuation triggers provide early warning of emerging risks and give the lender an opportunity to act before value materially deteriorates. Early intervention increases the range of constructive solutions available to preserve value and support an orderly resolution.

For investors, the relevant question is not simply whether a loan is secured by real estate, but how that security performs when the borrower's strategy falls short. The underwriting process should address that question before the loan is made, translating into appropriate leverage, pricing and structural protection. In CRE debt, the quality of the outcome is determined long before the downside arrives.