Page 1 of 1

Recent Advances from Safe Withdrawal Rate Research

Posted: Thu Aug 07, 2003 6:49 pm
by JWR1945

A Safe Withdrawal Rate is a prediction, an estimate, of how much one can withdraw from his retirement portfolio while satisfying well-defined constraints. Two typical constraints are to avoid running out of money during one's lifetime or to maintain the purchasing power of one's nest egg. Most of the time, but not always, results are specified in terms of maintaining one's purchasing power. That is, we usually adjust withdrawal amounts to match inflation. We generally speak in terms of the maximum that one can withdraw subject to constraints.

A Safe Withdrawal Rate is not an arbitrary estimate or prediction. It must be based upon historical evidence and it must be as objective as possible. Excluding a limited number of special cases, it is always the result of a mathematical calculation and it is taken from information available before the year in which it applies. The quality of a Safe Withdrawal Rate calculation is never defined in terms of what actually takes place in the future. A Safe Withdrawal Rate calculation seeks to identify the most likely outcome of future events. Just as probability theory is never judged by a single, actual outcome as when one flips a coin ten or twenty times, a Safe Withdrawal Rate calculation seeks to identify what is reasonably likely and what should be treated as unusual. It is never a guarantee.

It is critically important to introduce a similar term, which is quite different from a Safe Withdrawal Rate calculation. I shall call it the Historical Database Rate (HDBR). I am among many who have failed to distinguish between these terms in the past. Treating them correctly is critically important. Yet, even in my own work that I reference from this post, you will see me wrongly describe a Historical Database Rate as a Safe Withdrawal Rate.

For more about the definition of a Safe Withdrawal Rate, read this thread. Remember, though, that it has not yet been updated to include the new term Historical Database Rate.

A Historical Database Rate (HDBR) is the result of assuming various allocations and strategies to calculate what would have happened for a historical sequence of returns beginning at a specified year. This is the output for a single year in conventional Safe Withdrawal Rate calculators based upon historical sequences. A good example is the FIRECalc, which is very user friendly and which I have used frequently in my research. With FIRECalc, for example, if you set the portfolio lifespan to 30 years, the line for 1935 tells you what would have happened from 1935-1965. It is based entirely on the actual sequence of events that happened during 1935-1965, all of them occurring after 1935. In no sense whatsoever is it an estimate (or prediction) based entirely upon the information available in 1935.

You can meaningfully talk about a Safe Withdrawal Rate that includes some historical information. One could talk about the years 1935-1965, assuming that the first ten years matched the historical record exactly, with an estimate provided based solely upon information available in 1945. That would be the Safe Withdrawal Rate for 1945 for the 1935-1965 sequence. The initial ten years would influence the answer in a variety of ways, the most obvious of which is that it would establish the portfolio balance in 1945. With rare exception that balance would be different from what would have been predicted (for 1945) based on information available on or before 1935.

Results from Historical Database Rates are quite useful in helping us understand what has happened in the past. We extract information from those results and apply that information to produce our Safe Withdrawal Rate estimates. To the extent that we can identify credible cause and effect relationships, we improve our understanding greatly. We can rely directly upon those cause and effect relationships with far greater confidence than we could from looking at numbers alone.

Recent Advances related to Historical Database Rates

Earlier research had already established that portfolio volatility is the primary killer of retirement portfolios. From studies of Historical Database Rates came the famous 4% Safe Withdrawal Rate result. The key assumption was that the highest Historical Database Rate that was safe for each and every year in the database would still be safe going forward. This was described in the context that there has been a tremendous amount of variation of investment results in the past, both good and bad. It was assumed that all relevant factors affecting portfolio safety have been represented sufficiently in the past for making a reliable projection into the future.

From a series of events came proof that Historical Database Rates have been affected by valuations. Initially, it was found that P/E has had a slight predictive power. Later, thanks to BenSolar and his prodding of intercst, who made the actual calculations, it was found that Professor Shiller's P/E10 measure of valuation had reasonably good predictive power, albeit with a considerable amount of scatter. This relationship was strong enough to warn of a very real, serious danger to those who had retired in 2000 - at the peak of the bubble and its unprecedented valuation levels - and who were depending upon the 4% result.

BenSolar has often provided a link, as he did here, referring to the graph. ... 7080#p7080

Much more recently and looking forward to the Ed Easterling special event, I presented my methodology for determining Price-Adjusted Safe Withdrawal Rates. That methodology consists of ordering the years in the historical database according to P/E10. Then I examined Historical Database Rate results at various rates to estimate (crudely) the Safe Withdrawal Rate versus P/E10.
More recently still, hocus reminded me of one of my very first investigations. I had noticed that big gains or losses that went unanswered have had a profound effect on the safety of retirement portfolios. I had called for something reliable to be used as an indicator to predict such swings. In the more recent context, however, we had already identified a reliable indicator (P/E10). By marrying that early research with P/E10 and my Price-Adjusted Safe Withdrawal Rate methodology, I found an important anomaly in how the Historical Database Rate behaved. Results seen from 1881-1920 differ dramatically from those of later years. When we restrict our investigation to retirements beginning in 1921 and later, Historical Database Rates correlate tightly with P/E10. The large uncertainty previously encountered were the result of 1881-1920 anomaly. The behavior of Historical Database Rates versus P/E10 during 1881-1920 is very strange. It has no obvious explanation. It is not what one might expect if it were simply the result of randomness.

Excluding the 1881-1920 anomaly has allowed us to introduce valuation levels, as indicated by P/E10, into our examination of Historical Database Rates. As a consequence, we have been able to introduce them into our Safe Withdrawal Rate calculations as well.

Identifying Cause and Effect

Largely as the result of peteyperson's investigations into managing cash buffer accounts wisely, I have been able to focus on an especially bad period for starting retirements: the high valuation years of 1959-1973. What is striking about that period is that prices mostly went sideways or down (depending upon the starting point) but that dividend amounts were surprisingly consistent, generally growing with inflation. Looking at dividend yields throughout that entire period, I found that they were almost identically equal to the Historical Withdrawal Rates (assuming a final balance of zero). That shows an understandable cause and effect relationship during this extended period of high valuations and poor returns. Dividends explained it all. Only because the final balance was made to equal zero was there any selling of stock shares whatsoever. Such sales boosted the Historical Withdrawal Rates slightly, generally close to but less than 1%.

Based upon this observation, it was reasonable to describe a dividend-based Safe Withdrawal Rate strategy that was known to work at high valuations. The dividend yield determined one's withdrawal rate. One could boost his total withdrawals slightly, of the order of 1%, by selling some shares at (relatively) favorable prices. One's primary emphasis, however, should be to avoid selling heavily when stock prices are especially bad.

One important part of a dividend-based strategy is to focus on the quality of dividends. In those few cases that would have failed as found in the Historical Database Rate results, what happened was that dividends were cut sharply within two or three years. The dividend-based strategy would have worked based upon the new dividend amounts and their yields.

Extending the Range of Valuations

At this point I had identified what to do at times of high valuations but not at lower prices. The dividend-based strategy was not especially good at low and normal valuations. It was overly conservative. Most likely, stock prices would increase and it would be possible to increase withdrawals by more than 1% safely.

I tried the simplest possible adjustment. Surprisingly, it worked well. It is much better than I had expected it to be.

The process is to separate the two income streams, one from dividends and the other from selling shares of stock. I treat the dividend component as before. It is a steady stream of income that compensates for inflation reasonably well. To handle the price component, I scale the existing price to an equivalent at high valuation levels. Since we already know what to do to design a dividend-based strategy at high valuations, we do the same: we add something not exceeding 1% to the dividend yield to determine the Safe Withdrawal Rate at the equivalent high valuation level. Then we add what it would take (annualized over the lifespan of the portfolio) for stocks to rise from their existing levels to their equivalent at high valuation levels. ... 9492#p9492 ... 9495#p9495

It is far from perfect, but we can, at least, calculate a reasonably accurate Safe Withdrawal Rate using this procedure. We have to be careful to look into the quality of dividends, as we did before. But otherwise the procedure is straightforward and simple enough. We can, in fact, talk about the Safe Withdrawal Rate for 1935-1965 as opposed to the Historical Database Rate that actually occurred. The numbers are different.


Our understanding of Safe Withdrawal Rates has advanced considerably in the past few months. Much of what we have learned is sufficiently general that we can extend our knowledge far beyond limitations of allocating between an S&P 500 stock index fund and fixed income investment. We can even use our new findings to design withdrawal strategies from scratch.

Have fun.

John R.