"The investor's chief problem -- and even his worst enemy -- is likely to be himself."
Benjamin Graham, 1949

Human beings are hardwired for survival, not for investing. The psychological traits that helped our ancestors survive in tribes and navigate danger now often hinder our ability to build long-term wealth. At Canopy, we believe managing our own behavioural biases is as fundamental to our investment process as our rigorous quality assessment or valuation discipline.

Buy high, sell low!

Investment returns and investor returns are not the same thing. Markets have historically rewarded patient capital, but most investors fail to capture these returns due to poor timing decisions, a phenomenon known as 'the behaviour gap'. It occurs because investors systematically buy after prices have already risen and sell after they have fallen, transforming market volatility from a potential opportunity into a source of loss. Understanding and defending against this dynamic is crucial for long-term wealth creation, as even small behavioural penalties compound dramatically over time.

^{Source: The Behaviour Gap, Carl Richards.}

This pattern has been consistently documented in studies of investor behaviour. As noted in Chapter 4, DALBAR's long-running analysis found that over the last three decades, the average US equity fund investor has underperformed the S&P 500 by 3-4% annually, primarily because of buying high and selling low during volatile periods (2022). Morningstar's Mind the Gap study (2023) has reached similar conclusions, finding that the average dollar invested in funds earned a 10-year return to December 2022 that was 1.7% less annually than the average underlying fund.

Perhaps the most striking example comes from one of history's most successful fund managers. Peter Lynch managed the Magellan Fund at Fidelity Investments between 1977 and 1990, producing an annualized return of 29.2%. Yet it has been reported that the average investor in his fund made only 7% per annum (Jakab, 2016). Despite having access to exceptional investment performance, poor timing decisions cost investors the majority of their potential returns.

The behaviour gap tends to be most pronounced during periods of market exuberance or stress, as illustrated in Figure 2. In the lead-up to the GFC in 2007, equity fund inflows remained strong despite mounting economic risks. As equity markets declined through 2008, fund outflows accelerated dramatically, reaching their most negative levels during 2008-2009. As it turned out, this was actually a great buying opportunity. Crucially, outflows continued even after markets began recovering, with flows remaining negative through 2011 despite substantial market gains during this period. This timing pattern illustrates how investor behaviour often lags market turning points, leading to suboptimal entry and exit decisions.

^{Source: Investment Company Institute, Canopy Investors.}

When instincts backfire

To cope with the daily information overload, our minds rely on mental shortcuts called heuristics. While these shortcuts serve us well in many aspects of life, they can also work against us. Several behavioural biases are particularly problematic for investing:

• Confirmation bias causes us to seek information that supports our existing beliefs while ignoring contradictory evidence. An investor convinced of a company's potential might focus on management's optimistic guidance while dismissing customer complaints, executive departures, or deteriorating financial metrics.
• Overconfidence tricks us into believing we can predict outcomes with greater certainty than we actually can. This might manifest as an investor who has enjoyed success with a few stock picks becoming convinced they can time market cycles, leading them to make increasingly concentrated bets or using leverage.
• Loss aversion makes losses feel more painful than equivalent gains feel pleasurable. This asymmetry leads to seemingly irrational behaviour: holding losing positions for too long (hoping to 'break even') while selling winners too quickly (to 'lock in gains').
• Hindsight bias convinces us we 'knew it all along' after events unfold, preventing us from learning from our mistakes. An investor might remember being 'cautious' about a failed investment when their notes actually show they were highly confident at the time of purchase, preventing honest post-mortems that could improve future decision-making.
• Groupthink emerges when teams prioritise harmony over critical thinking. Investment committees might reach consensus not because the analysis is sound, but because challenging the prevailing view feels uncomfortable.
• Recency bias leads us to overweight recent events when making decisions. Investors often chase last year's best-performing asset class assuming recent trends will continue indefinitely. This backwards-looking approach systematically buys high and sells low, as yesterday's winners often become tomorrow's laggards.
• Herding makes us feel comfortable following the crowd, even when the crowd is heading toward a cliff. The Dutch tulip mania, the South Sea Bubble, and the housing bubble all shared a common thread: rational people making irrational decisions because 'everyone else was doing it'.

^{Source: Tom Fishburne.}

These biases often work in combination, amplifying their individual effects. Recency bias might lead an investor to chase last year's hot sector, while confirmation bias ensures they only read bullish research, and overconfidence convinces them to make an oversized bet. When a reversal eventually comes, loss aversion prevents them from cutting losses quickly, and hindsight bias ensures they learn nothing from the experience.

Our approach

The late Nobel laureate Daniel Kahneman reached a somewhat disheartening conclusion at the end of his influential book 'Thinking, Fast and Slow': there is little we can do as individuals to prevent behavioural biases from influencing our decisions. He considered that objectivity is easier when evaluating others' decisions than our own; "it is much easier to identify a minefield when you observe others wandering into it than when you are about to do so."

At Canopy, we have adopted team and rules-based processes to minimise individual biases, with our flat structure serving as the foundation:

• Institutionalized dissent: We have deliberately built structured debate into our process, recognising that it is often easier to spot flaws in someone else's reasoning than in our own. Every position in the Fund has a 'secondary analyst' whose primary task is to act as devil's advocate to the primary analyst's view. This structure reduces the prospect of groupthink and confirmation bias by forcing alternative perspectives into every investment discussion.
• Independent conviction ratings: Each team member maintains independent views on stocks, and these views directly impact portfolio construction. Ratings and comments are recorded and reviewed when performing post-mortems on positions, increasing accountability among all team members, reducing hindsight bias, and accelerating learning from both successes and failures.
• Pre-mortem analysis: Before taking positions, we routinely perform pre-mortems, where we imagine our thesis is incorrect a couple of years into the future and consider what might have gone wrong. This exercise reduces groupthink, confirmation bias, and overconfidence by forcing us to actively consider how our thesis might be wrong.
• Multiple valuation approaches: We use several valuation methodologies to avoid the pitfalls of relying on simple multiples or heuristics alone. Additionally, our valuations are expressed as ranges rather than point estimates, helping to reduce overconfidence, and acknowledging the inherent uncertainty in any valuation exercise.
• Disciplined selling framework: To combat loss aversion, we maintain strict sell discipline. We are particularly vigilant about selling stocks where there has been material 'thesis creep' - gradual changes to our original investment rationale that often signal we are rationalising poor performance. In our experience, these situations frequently become value traps. For high-quality companies, we more gradually reduce positions when valuation becomes the primary concern.
• Variant perception: When investing in any stock, we work to identify our source of edge and how our view differs from the market consensus. Since markets are typically efficient, the best opportunities arise when we can disagree with the consensus while maintaining strong conviction in our analysis.
• Systematic position sizing: We use a systematic tool that translates our quality and valuation assessments into suggested portfolio weights. As illustrated in Figure 4, higher-quality companies are allocated a substantially larger position size for any given expected return.¹ As these weights change daily with share prices (and less frequently with our fundamental assessments), we are prompted to dispassionately evaluate position sizes, helping to reduce the influence of recency bias and emotion.
• Decision quality over outcomes: We seek to evaluate the quality of our decisions rather than judging success solely according to outcome. This distinction helps us learn from both wins and losses by focusing on whether our analysis was rigorous and our reasoning sound, rather than results that may reflect luck or short-term market movements.

^{Source: Canopy Investors.}

While no process can eliminate mistakes entirely, ours is designed to reduce the risk of material missteps and support more disciplined position sizing. We accept that we will make mistakes, but by identifying potential pitfalls before committing capital - not after - we improve our chances of avoiding them. The same behavioural traits that once helped humans survive can undermine investment decisions. Our process has been designed to help our portfolio succeed in spite of them.

What happens when you ask artificial intelligence (AI) to predict how it might change the job of fund management - and then invite a leading AI researcher to respond? 

That's the premise behind this article, which explores how AI could reshape the investment management process within the next five years. 

We began by asking ChatGPT - powered by GPT-4-turbo - to generate a series of predictions, both optimistic and cautionary, about the future of fund management. 

Then we invited Prof. Sotirios Sabanis, who leads a team of AI researchers at the University of Edinburgh that's been collaborating with us at the Centre for Investing Innovation, to weigh in.

The result is an exchange between machine-generated foresight and human expertise - a glimpse into what the future might hold when algorithms and people routinely work side by side:

Portfolio construction gets a sidekick

AI's take: AI will assist fund managers by rapidly generating optimised portfolios based on real-time data, investor preferences, and risk constraints -- not replacing the manager, but acting as a tireless quant assistant.

Prof. Sabanis responds: Rapidly generated, optimised portfolios rely heavily - and will continue to do so - on finely-tuned optimisation algorithms, advanced statistical and stochastic modelling, and adequate computational resources. AI introduces a new level of automation by integrating the collection and processing of diverse data inputs - including some available in real time - with these core components. The result is the creation of information-rich decision-making tools that enable fund managers to gain deeper, faster and more granular insights. When developed with the right level of sophistication, these AI tools can act as supercharged quantitative assistants, offering fund managers unprecedented real-time analytical capabilities.

Sentiment as a signal

AI's take: Natural language processing will mine news, earnings calls, and social media to detect shifts in sentiment before they show up in prices. But beware: signal or noise?

Prof. Sabanis responds: Recent breakthroughs in natural language processing (NLP) have led to significant scientific advances, the full implications of which are only beginning to be explored across a wide range of application domains - including asset management - where early results have been promising. Sentiment analysis has seen a notable improvement in predictive power, driven by these NLP developments. However, sentiment cannot be fully captured through textual data alone. To enhance predictive accuracy and reduce noise in relevant datasets, complementary analytical tools are required. The integration of multimodal data - such as video, audio and text - enabled by emerging AI technologies, offers a promising path forward.

Forecasting gets a facelift

AI's take: AI models will enhance macroeconomic forecasting by identifying nonlinear patterns and hidden correlations -- though they may still struggle with black swan events and regime shifts.

Prof. Sabanis responds: There is growing evidence that AI models will significantly enhance macroeconomic forecasting by leveraging diverse, information-rich datasets and uncovering deeper correlations and interdependencies. A compelling example comes from recent advances in AI's predictive capabilities for inherently unstable systems, such as weather forecasting. These developments offer a glimpse into AI's potential in navigating complex, dynamic environments. However, accurately modelling extreme events well in advance remains a substantial challenge. The rarity of such events, coupled with their non-linear, non-stationary and non-periodic characteristics, limits the availability of training data and constrains the predictive accuracy - even for today's most advanced AI systems.

Human bias, meet machine bias

AI's take: AI may reduce some behavioural biases -- but introduce new ones. Overfitting, data drift, and model opacity could create false confidence in flawed outputs.

Prof. Sabanis responds: AI can act as a powerful behavioural counterweight - but only when its limitations are well-understood and carefully managed. In other words, AI holds significant promise as a behavioural stabiliser in financial decision-making - particularly through its capacity to reduce common cognitive biases such as overreaction, recency bias and confirmation bias. By anchoring investment strategies in structured data and statistical learning, rather than emotion or narrative, AI can enhance objectivity.

However, its effectiveness is not unconditional. AI systems are not neutral; they can introduce machine-driven biases rooted in their training data, algorithmic architecture and feedback loops. This creates a new class of risk - systemic, often opaque and harder to detect than human behavioural patterns. Risks such as overfitting, data drift and model opacity are particularly dangerous because they give an illusion of precision and reliability, often masked by high back test performance or technical complexity. If left unchecked, these flaws may propagate through portfolios, investment committees and client reporting chains - ultimately undermining the very trust that AI is meant to bolster. Hence, robust stress testing, continuous validation and model interpretability frameworks are essential safeguards to ensure that AI adds real value to the investment process - rather than introducing new risks or causing unexpected problems.

The rise of the 'explainable AI' arms race

AI's take: As AI becomes more embedded in decision-making, the pressure to explain its logic to clients, regulators, and boards will intensify. Transparency will be a competitive edge.

Prof. Sabanis responds: As AI becomes increasingly embedded in financial decision-making, explainability is no longer optional - it's a strategic imperative. Clients, regulators and boards will expect straightforward, credible answers as to how AI models arrive at key insights or investment decisions. Transparency will not only be a compliance requirement but a source of competitive differentiation.

Firms that can articulate how their AI systems work - what data is used, how decisions are made and where risks lie - will build stronger client confidence and demonstrate operational maturity. The future edge will lie not just in building powerful AI, but in making it explainable and trustworthy. This will require investment in interpretable models, documentation frameworks and governance protocols. Transparency serves not only as a means of mitigating downside risk, but also as a critical enabler of broader adoption, a catalyst for attracting institutional capital and a foundation for establishing long-term strategic credibility within an AI-driven financial ecosystem.