Hacker News Reader

Comment by Concrete | Hacker News Reader

My friends and I created a retirement calculator with many professional-grade capabilities. The model uses a 10,000 trial Monte Carlo simulation. We've made this pretty flexible, and have opened up the statistical settings for modification. We've also wired in "Fat Tails" for stock returns and added a "Black Swan" feature.

Each simulation requires generating millions of random variables and performing tens of millions of calculations. We understand requiring a sign-up seems like an unneccessary hassle, but we're scaling this on AWS and want to filter bot traffic due to the computational requirements. We apologize for this annoyance.

Happy to answer any questions you may have.

View on HN

Looks nicely done.

I'm curious why you chose to use the monte carlo approach? Other tools in familiar with use historical market data instead.

A lot of what I've read suggests that monte carlo doesn't approximate the market very well. The market is not random. It has well defined historical cycles (bull, bear, sideways). Also, issues like sequence of returns appear to have a large impact on portfolio longevity. How are you thinking about these sorts of issues?

Great questions. I strongly encourage folks to be careful with portfolio “back-testing”. There are a number of problems with using historical returns (or “boot-strap” sampling) for forward-looking analysis. The most conspicuous concern is the prevailing level of interest rates. Using historical stock and bond returns would entail annual performance of 10%-ish and 5%-ish for each asset class, respectively. In today’s environment, given high stock price levels (P/E ratios) and with interest rates fluctuating near historical lows, the consensus is that we should expect considerably lower returns for the foreseeable future.

For example, J.P. Morgan’s most recent capital market assumptions forecast a compound annual return of just 4.1% for large-cap U.S. stocks (e.g., S&P 500 Index) over the next 10 to 15 years. This is in stark contrast to the 10% to 12% historical returns pop-finance personalities often cite when setting expectations for future returns. For this reason, performing forward-looking analysis based on historical investment performance could result in grossly optimistic results and a false sense of security.

Monte Carlo simulation does in fact account for sequence of return risk through volatility and random number generation (but perhaps I misunderstood your point on that front).

As for the cyclical nature of returns, that’s a topic that continues to be debated. The historical behavior of markets in a developed country with enforced property rights, relatively strong institutions, and victories in two World Wars (e.g., the U.S.) is much different compared to other parts of the world. Maybe mean-reversion exists, or maybe we like to see patterns in data. In any event, we do plan to add customizable mean-reverting functionality in the near future.

Here’s some more info you might find helpful:

Studies of Market Behavior in Academia:

https://www.honestmath.com/learn/academia

Perspectives on how little we actually know, and how little historical returns might actually tell us:

https://www.honestmath.com/learn/nobody

Some more relevant thoughts:

https://www.honestmath.com/learn/simulation-probabilities-ar...

Reply on news.ycombinator.com