4-Minute Read

716 words

Recently, FIRE (Financial Independence and Retire Early) has become quite popular. People probably don’t like going to work. The source of income for not working is the 4% Rule, but this is actually a simplified result with dnagers.

The 4% Rule

The 4% Rule was proposed by William Bengen in 1994 “Determining Withdrawal Rates Using Historical Data”. It involves investing retirement funds in stocks and bonds, and withdrawing 4% annually for living expenses. Because financial assets will grow, there is no need to worry about running out of retirement funds. This concept is not wrong, but it ignores many details.

The simulation method assumes that client X retires in a certain year. In the first year, 4% of the investment portfolio market value is withdrawn. After that, the withdrawal amount is adjusted annually for inflation. If the simulated client retires in any year between 1926 and 1976, how many years can the investment portfolio last for over 50 years without being depleted to zero? Because 1976 + 50 = 2026, beyond the year 1994 when the author wrote this article, the author assumes that the returns after 1994 have not changed significantly. The article does not describe in detail how he assumed this. I assume that the annualized returns used are the same as the historical data, 10.3% for stocks, 5.1% for bonds, and 3% for inflation. Therefore, with a 60/40 stock/bond allocation, rebalanced annually, the inflation-adjusted annualized return is 5.1%. Withdrawing 4% should be sustainable for over 30 years.

Assumptions of the 4% Rule

  1. No taxes are paid on withdrawals.
  2. The investment portfolio only needs to last 30 years.
  3. Future returns, inflation are assumed to be similar to 1976 to 1993.

If taxes need to be paid on withdrawals, then the withdrawal rate has to be higher, or the usable cash after tax is lower. The current average life expectancy in Taiwan is around 80 years old. So for those looking to retire before 50, the 4% rule may not apply. There are variances in life expectancy between individuals, and average life expectancy in Taiwan may exceed 80 years in the future due to technological advances. Also, people who are financially well-off enough to consider early retirement will likely have above average health!

In-sample Conclusions

The third point is the most dangerous - assuming future returns and inflation will be similar to 1976 to 1993. This article made a statistical mistake by deriving all numbers “in-sample” without out-of-sample testing. If future returns or inflation differ from the sample data, the original conclusion becomes invalid. Tax costs and longer time horizons can be simulated again, but a flawed method will only lead to flawed conclusions no matter how many simulations are run.

Out-of-sample testing

In machine learning, the sample used to measure performance must be separate from the training sample, otherwise the measured performance does not reflect real-world performance. Imagine an extreme example: I can find a region’s lottery numbers for 5 consecutive draws that match the next day’s stock price movement of a certain company’s stock in a certain country. Clearly there is no causal relationship between the two, and using the next draw to predict the stock price movement the day after makes no sense. Even if it predicts correctly for all 5 draws in-sample, out-of-sample accuracy can be expected to be no better than random guessing.

Similarly, many funds and ETFs tout too good to be true backtested returns before launching, but those are all in-sample data. Parameters can be tuned to beat the market return in-sample. It have no predictive power to the post-launch returns (out-of-sample data), which is why the regulator requires the disclaimer “past performance does not indicate future results”.

How to Retire Properly

I’m not saying the 4% rule cannot be used. I’m saying based on the reasoning process behind the 4% rule, we cannot conclude this method will still work for the next 30 years. Financial assets is a good place to store value, keeping financial assets and making periodic withdrawals as in the 4% rule is still a reasonable method. A better method is having sources of income not dependent on financial assets even without a full-time job, or maintaining employability (tutoring, consulting etc), as well as having good insurance.

Reference

“Determining Withdrawal Rates Using Historical Data”

comments powered by Disqus

Recent Posts

Categories

Tags