This post is the last of three on inflation during retirement:

- Cost of living adjustment (COLA). Analysis of the 0% COLA for Social Security in 2015.
- Better ways to measure what this year's inflation really was.
- Predicting inflation next year, to better forecast your budget (this post).

## Don't Look Back

Question: What bias affects both the cost-of-living adjustment (COLA) from the Social Security Administration (SSA) and the oft-cited 4% Rule for retirement income?

Answer: Both use last year's inflation. They look back when they should be looking ahead.

Admittedly, inflation in recent months is somewhat correlated with inflation a year from now. Since 1914, the Consumer Price Index (CPI-U) in any given month has had a correlation of 0.54 with inflation 12 months ahead. Over the last 30 years, however, the correlation has declined to 0.21, a value so low that it's next-to-worthless for making predictions.

By looking backward to predict forward, both SSA and the 4% Rule are, in effect, adjusting in arrears. SSA waits until inflation has happened, and then modifies payments as of January the following year. Similarly, the 4% Rule advises that late in the year, you should find the past year's inflation and increase your retirement spending by that amount in the year to come. In both cases, the inflationary train has left the station and reached the next stop on its journey. You can catch the next train, but you'll always be one station behind.

Answer: Both use last year's inflation. They look back when they should be looking ahead.

Admittedly, inflation in recent months is somewhat correlated with inflation a year from now. Since 1914, the Consumer Price Index (CPI-U) in any given month has had a correlation of 0.54 with inflation 12 months ahead. Over the last 30 years, however, the correlation has declined to 0.21, a value so low that it's next-to-worthless for making predictions.

By looking backward to predict forward, both SSA and the 4% Rule are, in effect, adjusting in arrears. SSA waits until inflation has happened, and then modifies payments as of January the following year. Similarly, the 4% Rule advises that late in the year, you should find the past year's inflation and increase your retirement spending by that amount in the year to come. In both cases, the inflationary train has left the station and reached the next stop on its journey. You can catch the next train, but you'll always be one station behind.

## Look Ahead

A better idea, clearly, would be to use an accurate predictor of next year's inflation, something better than simply assuming that the future will mimic the recent past. How might one do that?

One idea is to use the median (middle) estimate of experts whose livelihood depends on precisely forecasting future inflation. As it happens, this method is easy to implement and reasonably accurate. The Philadelphia branch of the Federal Reserve Bank conducts a survey of professional forecasters and publishes the results quarterly. Over the last 30 years, the correlation between this forecast and the year-ahead value of the broad CPI-U measure of inflation has been 0.52. As shown by the chart below, the survey tends to ignore brief divergences, up or down, and predict something like an average trend.

One idea is to use the median (middle) estimate of experts whose livelihood depends on precisely forecasting future inflation. As it happens, this method is easy to implement and reasonably accurate. The Philadelphia branch of the Federal Reserve Bank conducts a survey of professional forecasters and publishes the results quarterly. Over the last 30 years, the correlation between this forecast and the year-ahead value of the broad CPI-U measure of inflation has been 0.52. As shown by the chart below, the survey tends to ignore brief divergences, up or down, and predict something like an average trend.

Indeed, the survey correlates extremely well (0.92) with future core inflation, as measured by the CPI without its volatile food and energy components. In effect, the forecasters, although asked to predict the broadest measure of inflation, CPI-U, are actually estimating core inflation. That's good because, as argued previously in this series, core inflation is a good way to manage a retirement budget. Our Safe-Payout calculator therefore uses the survey forecast as the recommended COLA for those who set their retirement budget with the "collared inflation" method (a safe variant of the 4% Rule).

Is it possible to do even better than the survey forecast? After all, the Philly Fed is not Lake Wobegon, "where all the children are above average." In the survey, some forecasters must do better than others, but we aren't told who they are or whether their superior performance persists over time. Cherry-picking the best forecasters would be perilous, admittedly, because statistical theory says they will tend to regress, doing less well in the future than they have in the past.

With that precautionary observation in mind, take another look at the chart above. It offers an additional forecast based on an econometric model from academic research. Shown by the red line in the chart, the model's correlation with the broad CPI-U metric has been 0.83 since 1978. The model is a simplified version of the so-called "triangle model" of Robert J. Gordon, an economist at Northwestern University who developed the model and has applied it to data for a 51-year period from 1962 to 2013.

In the simplified form shown here, the model says future inflation depends on:

Is it possible to do even better than the survey forecast? After all, the Philly Fed is not Lake Wobegon, "where all the children are above average." In the survey, some forecasters must do better than others, but we aren't told who they are or whether their superior performance persists over time. Cherry-picking the best forecasters would be perilous, admittedly, because statistical theory says they will tend to regress, doing less well in the future than they have in the past.

With that precautionary observation in mind, take another look at the chart above. It offers an additional forecast based on an econometric model from academic research. Shown by the red line in the chart, the model's correlation with the broad CPI-U metric has been 0.83 since 1978. The model is a simplified version of the so-called "triangle model" of Robert J. Gordon, an economist at Northwestern University who developed the model and has applied it to data for a 51-year period from 1962 to 2013.

In the simplified form shown here, the model says future inflation depends on:

__Inertia__, or the tendency of core inflation to persist. While Gordon uses a weighted average of the previous five years of inflation, my analysis of data since 1978 found equally good results by simply using the most recent 3-month average of the trimmed mean PCE measure of inflation.__Demand for employment__, as measured by the gap between short-term unemployment and a "natural rate." If, say, only 3% of the labor force has been unemployed for six months or less, and the natural rate is higher than that, then jobs are so easily found and quickly filled that upward pressure on wages and prices will soon push inflation higher.__Supply of critical goods__, which can cause inflation to fall or rise if temporary supply-shocks boost or diminish the availability, and thus affect the price, of those goods. In the simplified model, I used just one value, which had statistical significance for the period 1978-2015: an average of the energy component of the CPI for the previous quarter.

## Good Advice

- To budget for inflation of consumer prices in 2016, a retiree could prudently use the mid-point estimate of professional forecasters, which currently stands at 2.04%.
- An alternate forecast from a good econometric model predicts that in the next three months, inflation is likely to be mild, about 1.15%, but it could rise to 2.24% by this time next year.
- A caveat for any prediction is that future inflation will depend on some things we know now (like today's rates of inflation and unemployment) and on some events we can't know in advance (like the effect of currency exchange rates on imported goods and the market prices of oil and gas). For this reason, I plan to add a new calculator that will provide updated, quarterly forecasts of inflation, using all the methods outlined in this post.