AI

How Do Tech Bubbles Pop?

by

Quiz!

How long did it take the Nasdaq to reach a new peak after the 3/10/2000 peak?

  1. It kept on going up – the Nasdaq doesn’t decline given the solid technologies produced by the companies in it
  2. 3 years
  3. 6 years
  4. 10 years
  5. 14 years
  6. It is unknown yet – it is still recovering.

How Do Tech Bubbles Pop?

On 1/27/2025, Nvidia and Broadcom stocks declined 17%. Why would the stocks of two of the leading AI companies in the world get hit so hard? This was in response to Chinese DeepSeek’s AI competing with ChatGPT, while its API access (access for programmers) is offered at less than 1/50th (2%) of the price of the ChatGPT API.

This is a common step in tech bubbles, going at least as far as the Roaring 1920’s (commercialization of the radio, TV and more) that were followed by the Great Depression. Here are typical phases of tech bubbles, from formation to popping:

  1. A new technology is introduced offering great value.
  2. As the technology gets demonstrated and improved, adoption increases, providing great profits to the companies offering it.
  3. The stock prices of companies offering the technology increase given their profitability.
  4. At some point, many investors buy the stocks given past returns and the future promise, while ignoring valuations, such as Price relative to Earnings and Book Value (P/E and P/B). That is when the gains form a bubble and can be seen with a spike in valuations.
  5. In the meantime, two effects happen:
    1. Economies of scale lower the cost to produce products with the technology.
    2. Competition, with a focus on efficiencies, offers the technology at much lower costs.
  6. As competition increases and costs go down, the entire industry lowers the prices. This includes the original companies offering the technology.
  7. With much lower prices, profitability of the producers of the new technology goes down, and the prior valuations become disconnected from reality, leading them to decline.
  8. That leads to a crash in those technology stocks, all while offering a very valuable service.
  9. While there is a logical progression, the timing isn’t easy to identify in advance.

Was 1/27/2025 the peak of the current tech bubble? It is tough to say, before reaching the depths of the decline. In 2022, we got a false peak with the tech-heavy Nasdaq declining a mere 33% (vs. the 78% decline in the dot-com crash that started in 2000). After 1 year, it bottomed and reached new highs less than 2 years later.

Once the real decline starts, how long can it take to recover? As typical, there are no guarantees in investing. Having said that, there has been a correlation between the excess valuations (relative to the typical) and the length of the decline.

  1. In the 2000’s the S&P 500 took over 10 years to recover, with a 30% decline a decade after its peak.
  2. In 1989 Japan reached significantly higher valuations that led to a decline to recovery of over 30 years, given much higher peak valuations.
  3. Fast-forward to 2025: The P/B (or Price/Book) of the S&P 500 on 1/27/2025 was slightly higher than the highest point in 2000, so another “lost decade” shouldn’t surprise us.

Quiz Answer:

How long did it take the Nasdaq to reach a new peak after the 3/10/2000 peak?

  1. It kept on going up – the Nasdaq doesn’t decline given the solid technologies made by the companies in it
  2. 3 years
  3. 6 years
  4. 10 years
  5. 14 years [The Correct Answer]
  6. It is unknown yet – it is still recovering.

Explanation: The Nasdaq declined by 78% over a span of 2 years and 7 months, followed by a long road to recovery, with a total of 14 years peak-to-peak.

Disclosures Including Backtested Performance Data

My Personal Experience with the Recency Bias

by

Quiz!

Which diversified investment looks more appealing?

  1. 15% average growth per year over the past 10 years, up from a long-term average of 10% per year.
  2. 5% average growth per year over the past 10 years, down from a long-term average of 10% per year.

Say that after 2 extra years, the faster growing investment continued performing better than 10% per year. Which would you choose now?

  1. The first one.
  2. The second one.

My Personal Experience with the Recency Bias

What is the Recency Bias? It is making decisions based on recent events, with the expectation that they will continue.

How can the Recency Bias hurt investors? Most investments are cyclical, while the recency bias assumes no cycles. Common harm is buying high after unusual gains or selling low after unusual declines. When done repeatedly, it can lead to long-term underperformance of a simple buy-and-hold strategy.

Are there less obvious cases of Recency Bias hurting investors? Yes. Many investors are disciplined enough to hold onto their investments at low points, but they may wait for gains before investing new money. Missing a 1% or 2% gain is nearly harmless. But some investors wait for more and more evidence. Once they see (and miss) 20% or 30% gains, some wait to buy at a dip, and some wait for more evidence of gains. Only after seeing 50% to 100% gains, some feel that the gains are here to stay, and invest after missing out on huge gains. The damage is far worse than simply missing gains, leading to a negative snowball. The delayed investment hurts their personal returns, they think that their investments are worse than reality, so they stay less committed to them, hurting their returns even further, cycle after cycle.

Did I ever experience the Recency Bias? Yes & no. When trying to think about the likely returns of an investment in the next 10 years, I know that it’s likely to be different than the past 10 years, given studies of investment cycles and valuation measures. But, in anomalous times, where a cycle gets stretched longer than usual, I am tempted to temper my expectations for the next leg of the cycle. I recognize that real life works the opposite – the longer we have an anomaly, the stronger the reversal tends to be. Examples of my recency bias:

  1. When looking at the raw data, it is rational to expect the S&P 500 to lose value over the next 10 years. But the recency bias leads me to believe it may get low positive returns.
  2. When looking at the raw data, it is rational to expect non-US Value (low Price/Book) stocks to enjoy unusually high returns over the next 10 years. But I catch myself sometimes expecting only average returns.

How damaging can the Recency Bias be? The examples I gave right above are not too harmful. They don’t lead me to make decisions that are opposite of rational, so I can live with them. The harm comes from an expectation opposite of rational, that leads to decisions that are very likely to fail. Here are examples:

  1. Expecting the S&P 500 to average more than 10% per year in the next 10 years, or even 6% or 8%.
  2. Expecting low interest rates in the next few years.
  3. Expecting AI-focused companies that reached extreme valuations to significantly outperform the rest of the market in the next 10 years.

How do I avoid big harm by the Recency Bias? I base my expectations based on a combination of:

  1. Full cycle, long-term behavior.
  2. Logic.
  3. Valuations (e.g. Price/Book) today relative to typical in the past.

Quiz Answer:

Which diversified investment looks more appealing?

  1. 15% average growth per year over the past 10 years, up from a long-term average of 10% per year.
  2. 5% average growth per year over the past 10 years, down from a long-term average of 10% per year. [The Correct Answer]

Explanation: High growth diversified investments tend to be cyclical, with reversals being more common than not after 10 years.

Say that after 2 extra years, the faster growing investment continued performing better than 10% per year. Which would you choose now?

  1. The first one.
  2. The second one. [The Correct Answer]

Explanation: When above/below trend continues beyond 10 years, reversals continue to be the more common case, with greater odds and magnitude.

Read this month’s article to find out what leads people to pick the other option for both questions.

Disclosures Including Backtested Performance Data