The Duration Gap: Value-at-Risk, Interest-Rate Swaps and the 2026 Losses of Chile’s Most Conservative Pension Fund

In the world of retirement savings, there is a fundamental tension between safety and growth. For people nearing retirement, the goal is usually to protect what they have earned, avoiding the wild swings of the stock market. In Chile, this protection comes in the form of a specific type of pension fund known as Fund E. It is designed to be the most conservative option, investing primarily in fixed-income assets like government bonds, which are generally considered stable. To manage the risk of these investments, fund managers use mathematical tools to predict how much money they could lose in a bad week or month. One of the most common tools is called Value-at-Risk, or VaR. Think of VaR as a weather forecast for money: it looks at how stormy the financial skies have been in the past and uses that history to predict the chance of a storm tomorrow. If the past few years have been calm, the forecast suggests the risk is low. However, this system relies on a crucial assumption: that the portfolio being measured today looks very much like the portfolio that existed in the past. If the portfolio changes its nature quickly—adding new, hidden risks that the past data never saw—the forecast can become dangerously wrong.

This is exactly what happened in Chile between 2023 and 2026, a period that revealed a blind spot in how these funds measure danger. Researchers analyzed the performance of Chile's pension funds, focusing on the conservative Fund E, which is the default choice for those closest to retirement. They discovered that while the funds appeared calm and safe on paper, their managers had quietly built up a massive, hidden exposure to interest rates in the United States. Specifically, the funds had entered into complex financial agreements called interest-rate swaps. These agreements act like a leveraged bet: if US interest rates fall, the funds make money; if rates rise, they lose. Between late 2022 and early 2026, the funds increased their sensitivity to US rates from a tiny fraction of a year to over seven years. This means that for every small move in US rates, the value of the fund moved as if it were holding a bond with a seven-year lifespan. When US interest rates began to rise sharply in 2026, the funds suffered a devastating blow, losing about ten percent of their real value in just nine months. This was the second-worst result on record for this conservative fund, and the losses were almost entirely caused by this exposure to US rates.

The most striking finding of the study is that the standard safety checks failed to sound the alarm. The researchers tested the standard Value-at-Risk models that the funds used every day. These models looked at the daily price movements of the funds and calculated the risk based on recent history. Because the funds had built up their risky positions during a period of financial calm, the recent history showed very little volatility. Consequently, the models reported that the risk was low and within normal limits. In fact, the models passed their own tests; the number of days where the funds lost more money than predicted was within the expected range. The researchers found that the loss was not a series of sudden, shocking crashes, but a steady, persistent drift downward as rates climbed. Because the daily volatility remained low, the standard tools saw a calm sea, even though the ship was drifting toward a waterfall. The models were looking at the wrong thing: they were measuring how volatile the fund had been, not what the fund was actually exposed to.

To understand the true danger, the researchers created a different kind of measurement. Instead of relying solely on the fund's past price movements, they looked at the current structure of the portfolio and asked a hypothetical question: if this exact portfolio had existed during the turbulent market moves of the last four years, how much would it have lost? This "exposure-adjusted" measure told a completely different story. By mid-2026, just before the worst losses hit, this new measure showed that the risk was more than double what the standard models predicted. While the standard tools suggested a safe one-month risk of about 3.5 percent, the exposure-adjusted tool warned of a 7.4 percent risk. This warning was accurate; in September 2026, the fund lost 5.1 percent in a single month, a loss that the standard models had failed to foresee but the exposure-based measure had correctly anticipated. The study also examined whether different fund managers had different levels of risk. While the managers with the highest exposure did suffer the largest losses, the small number of funds meant the researchers could not statistically prove that the exposure level was the sole cause of the difference, though the pattern was clear.

The episode serves as a powerful lesson about the limits of relying on history to predict the future. The researchers concluded that the standard Value-at-Risk models did not fail because they were broken, but because they were answering the wrong question. They successfully described how the portfolio had behaved in the past, but they could not see that the portfolio had changed its nature. When a fund builds up a massive new exposure quickly and quietly, the history of its returns does not yet reflect that new reality. The gap between what the fund was actually holding and what its history suggested it held became so wide that the standard safety nets were useless. In response to this kind of risk, Chilean regulators had already introduced a new rule in April 2026 that limits derivatives based on stress tests rather than past volatility. This new approach forces managers to look at what the portfolio is exposed to right now, rather than just how it has behaved before. The study confirms that for long-term investors like pension funds, understanding the current shape of the bet is far more important than trusting the weather report of the past.