Is your asset really outperforming? The basic return math you should know

[crixus123]

Raise your hands if you've heard some version of this:

I noticed fund A / stock B has outperformed Nifty over last N months / years.

This would be a determining factor for the investor to eventually go with fund A or stock B.

For instance, someone in our Discord had asked yesterday, since Tata Resources & Energy Direct Growth fund has outperformed Nifty in the last 3 months, is it a good time to enter this fund.

Even SPIVA reports have given index chasers the kind of peace of mind that in last N years, x% of active funds have failed to beat the index. Therefore, picking an active fund would mean x% chance of failure, going forward.

In this story, x is usually much higher than 50, across different long term durations.

This doesn't compute! It's wrong. All kinds of wrong. Even simple mathematical formulation would show that's not how it works at all.

Let's look at some real data.

Starting from January 1st 2019, Axis Bluechip Direct Growth has outperformed UTI Nifty Index Direct Growth, till yesterday.

This is a well-defined time period (1st Jan 2019 - 7th May 2021).

From Valueresearch fund page of Axis Bluechip Direct Growth

Clearly, if someone had invested a lumpsum amount in Axis Bluechip direct growth on 1st Jan 2019; they'd have done better than investing same amount in UTI Nifty Index fund on that same day. No question on that.

But what if you're more of an SIP investor? Which most average retail investors would be, because of monthly income and investing in the markets.

An SIP of 10k / month, starting from same 1st Jan 2019 in both funds would behave very differently.

SIP in Axis Bluechip Direct Growth is valued at 375,055 INR, at 23.2% p.a.

SIP in UTI Nifty Index Direct Growth is valued at 375,211 INR, at 23.3% p.a.

As you can see, for an SIP investor investing in UTI Nifty index fund would've done better than investing in Axis Bluechip direct growth, even when Axis Bluechip has outperformed UTI Nifty index fund.

So, you're saying index funds are better?

No, not even remotely the point I was making.

All I'm saying is over same time duration, if asset A beats asset B in point-to-point returns, that doesn't mean investors in asset A would see better returns in their portfolio, than investors in asset B.

It might often even be the opposite.

For instance, you could be able to show me situation (a given time period, a set of transactions; where active fund in a portfolio does better than index fund, even when the fund has failed to beat benchmark over same time period) where the opposite is true.

Well, duh! Past performance is not indicative of future performance

No, look closely. It's over same time period.

Not different time periods. Same start date, same end date.

You are comparing lumpsum returns vs SIP returns. We know these would be different. What's new about it?

Keep reading, it's not just lumpsum vs SIP. Not saying SIP is better or worse.

All I'm saying is do not make a guess about SIP returns based on lumpsum returns over a given period. Not only these would be different, they might even be opposite.

But what does this mean? Why does this happen?

Take a close look at this diagram, of two assets' price movement over a certain period.

Asset A has beaten Asset B over a certain period of time (between

t0

and

t1

), but as you know from our discussion above, SIP / DCA investor in asset B would've done better.

[NOTE: DCA stands for Dollar Cost Averaging. It's firangi version of SIP]

If you intuitively think about it, asset B gives more opportunities to buy at relatively lower prices.

But asset A keeps purchase prices higher as well.

And now, shift the start date from

t0

to closer to (but before)

t1

- you'd see that over this smaller time period, asset B moves up more than asset A.

Meaning, it outperforms asset A in a smaller time period later down the line, with same end date as before.

In our example of Axis Bluechip vs UTI Nifty earlier, if you change start date to 23rd March 2020, it becomes apparent.

Notice how the tables have turned!

To summarize:

• buy more when asset is undervalued or underperforming another asset
• wait for the asset to start outperforming the other asset

This is same as saying asset returns are not portfolio returns, which is a well known general principle.

So, instead of looking at outperforming funds, should I now start looking at underperforming funds?

No, not at all.

It's not the underperformance that made the asset B better. That's just one part of it.

It's the subsequent outperformance.

If anything, this should not be taken as a way to make better asset selection.

Rather look at it as even when you invest in assets that can outperform another asset and get it right, you might not be seeing any outperformance in your portfolio.

The timing math, just got much more complex.

Transaction Pattern

Return one would see in their portfolio, would depend on transaction dates, relative amounts in each transaction.

Just like we compared lumpsum and SIP; we should also look at actual transaction pattern of each investor.

Most investors won't be running a steady SIP of same fixed amount every month.

Rather, they would do some of these as well:

• step up SIP every time their incomes increase
• stop / skip some SIP
• invest more during red days, trying to time the market
• switch to other funds or add new funds

Each of these, give birth to a new transaction pattern. A set of transactions / cashflows, in a given asset / set of assets.

It'd be unique to each investor.

And therefore, the returns they'd be seeing would also be unique to each investor, and in no way related to the returns next generation of investors would be seeing in the aggregator portals.

Next time you're wondering why is a certain finance blogger holding a dud fund like Quantum Long Term Equity in their portfolio, entertain the thought that they might be seeing something very different in their portfolio.

 

[rae46]

@crixus123 What's the actionable?

What should I do?

 

[crixus123]

@rae46 Good question. To begin with, don't pick funds based on past outperformance.

Outperformance in point-to-point returns don't mean investors were able to experience the same in their portfolio.

This is assuming the outperformance continues and you were right on your money, with your pick.

Second one - when benchmarking your portfolio, simulate transactions in an investible version of the benchmark(s).

I've seen plenty of people go I've been investing for 6 months and made 60%, Nifty has 40% return within that period of time, therefore I'm next Buffet / Munger.

Ideally, he shouldn't compare returns of portfolio with point-to-point returns of a benchmark. Instead, create same purchase transaction / sell transactions on same dates, for same amounts in Nifty / any other benchmark, to compare.

 

[rae46]

@crixus123 Makes sense but Too much processing for my little brain.

When I invest blindly people say analyze

When I analyze people say hire a pro,

When I hire a pro people say pro has no skin In the game

 

[esy]

@rae46 Haha, ultimately you should feel content about your investments and the return they provide you and not go trying to min-max too much.

 

[rae46]

@esy No one is content in life na... That's the other problem

 

[resjudicata]

@crixus123 Can you share an example please. For eg. if you have a fund of 10K per month.

 

[mariah36]

@crixus123 Very well illustrated. Thank you!

Edit: This is why 1/3/5 year trailing returns are simply meaningless. The industry needs to move beyond this.

 

[dame]

@crixus123 Long term rolling returns is the gold standard..

 

[crixus123]

@dame But even with long term rolling returns, you're simulating a transaction pattern that's not yours, right?

And the post is about showing different transaction patterns give rise to different returns & final corpus size; from same asset, over same time period.

 

[nischansr]

@crixus123

But even with long term rolling returns, you're simulating a transaction pattern that's not yours, right?

Rolling returns charts are unaffected by transaction patterns - assuming you'd have picked one of the two funds you're comparing at any point where you had money to invest. Infact they're so simple to read that I am surprised they aren't used isn't used more (afaik, only rupeevest provides it).

The trouble people run into comes from the fact that many of them fail to see that you cannot use a 1 year rolling chart if you want to know what happened over a span of 10 years (for example). A N year rolling returns chart simply shows the point to point returns for two dates N years away. People confuse this with the time window of the plot/chart it seems to me.

So for finding out SIP returns it's best to simply use a SIP returns calculator (or understand this gotcha when eyeballing the rolling returns charts).

Excellent post btw - I wonder how you even find enough time in your life to write all of this down

This is something sooo many people don't seem to grasp.

 

[crixus123]

@nischansr

Rolling returns charts are unaffected by transaction patterns - assuming you'd have picked one of the two funds you're comparing at any point where you had money to invest.

Rolling returns would obviously be affected by it.

Here's an SIP in UTI Nifty Index fund Direct Growth from 1st Jan 2015, for 6+ years.

Now switch that date to 28th of month, and returns slightly improve.

Now consider same person investing 20k / month from 1st Jan 2019 (100% step-up, after 4 years of investing in the same fund)

As you can see, with deviation in transaction patterns, returns at the end of 6th year change, often quite so much. So if you're doing just rolling returns for 6 year periods, even between 2015-2021, there are infinite possibilities with varied range of results.

I feel I should've clearly defined what a transaction pattern is.

A transaction is defined as follows: It's a tuple of transaction date, normalized amount, and type of order.

For example:

(date='21/01/2021', normalized_amount=1, type_of_order='buy')

Transaction pattern is collection of all transactions such as this.

A user's CAS of historic transactions is a de-normalized (all amounts can be divided by a common divisor to normalize those) would be an example of transaction pattern.

This would be unique to each user, and correspondingly returns would also be unique to each user; even when two investors are investing in same asset with same start date and end date.

Basically, there are infinite possibilities.

I'd ideally prefer if there's a way to model transaction pattern of an average investor, using some kind of probability distribution.

Infact they're so simple to read that I am surprised they aren't used isn't used more (afaik, only rupeevest provides it).

I don't think rolling returns are as useful.

It has high self-correlation and creates the illusion that you're looking at lot of independent data sets, but in reality you're not.

I use rolling returns as a way to understand how good or how bad it can get. Basically, expectation setting with equity investing is best done with rolling returns.

However, I still think services should provide it. If I'd to guess, it's about not enough ROI, so not a priority. What does a business stand to gain providing it en masse, when most people would've to be educated about what it is?

So until there's a service out there that's popular and influencing enough to make it main-stream, no one else would jump to put it on their fund pages / screener pages.

As you'd imagine, things like these is why we started the wiki project.

 

[nischansr]

@crixus123 I sense that we are comparing/talking about different things. (I was using the phrase transaction patterns to mean something else.)

In my comment I am using rolling returns as a means of comparing two funds, while you are more interested in the performance of an individual given a fixed instrument (and I agree with the conculsions of your original post).

It has high self-correlation and creates the illusion that you're looking at lot of independent data sets, but in reality you're not.

This is not the right way to look at it.

One should not summarize/clump the rolling returns charts and simplify it to a distribution because it takes away an important piece of information (imo) - temporal consistency (and magnitude) - provided by this measure. Rolling returns charts are not very useful in isolation - they are really useful when comparing two funds or a fund and an index. Knowing the distribution of returns is less useful (for my purpose in using a rolling return chart - fund picking) than knowing the temporal performance.

The different points in a rolling return chart are every bit as independent/dependent as a regular NAV/index value chart. After all, one is just a derivative of sorts of the other. There is no new information in the rolling return charts - just some information presented differently.

If there is self correlation in the rolling return charts then it exists because of the performance of the underlying index/fund. The velocity chart of a car (NAV/index value charts) has the same information thats present in its acceleration chart (rolling returns chart). To be more pedantic the velocity chart also has some extra information - the constant, in this case the initial velocity.

 

[mariah36]

@crixus123 Agree. Rolling returns show performance of lumpsum investments, not of SIPs. This is an important distinction.

 

[crixus123]

@mariah36 You can even have SIP rolling returns. But even then, which date to pick?

And as I’ve mentioned above, even investing solely via SIP is rare. People change up amount and dates, all the time.

 

[nischansr]

@mariah36 No, this isn't true (or more like only half true). A rolling result chart shows point to point returns so it doesn't matter what you did - sip or lumpsum.

But if you can find a fund that does better than another fund in the rolling returns chart for most of the time, then the SIP returns are almost certainly going to reflect the same story.

If you just want to check sip returns, do an SIP check (particularly since volatility can benefit SIPs)

Rolling returns charts are often misunderstood/misused but when used well are excellent tools for an investor.

 

[gail76]

@nischansr I know your post is almost a month old but I could not resist commenting on it anyway.

You do understand that rolling return is exactly same as a specific period trailing return with just extra data points because of start and end date pairs being multiple? If a trailing return can be impacted by transaction pattern why would not an individual result of a rolling returns set?

Also, even though rolling returns is better than just looking at trailing returns it does not much help in picking one fund vs the other. The starting point of overall date range impacts the final outcome a lot.

 

[nischansr]

@gail76

You do understand that rolling return is exactly same as a specific period trailing return with just extra data points because of start and end date pairs being multiple?

I'm not sure what that's supposed to mean.

N-year Rolling returns is the set of all point to point returns of an instrument where two points of measurement are N-years apart (preferably arranged in chronological order)

Rolling returns for a fund is not one number (like xirr, or sip returns in the last N years for a specific kind of sip etc), it's a sequence of numbers. Most of the misunderstandings happen because people use various averages to take that sequence and derive a single number out of it.

 

[gail76]

@nischansr A 5 year trailing return today is the return from 7 jul 2016 to 7 jul 2021. 5 year rolling return is nothing but calculating the exact same thing for different date pairs over and over again. So if transaction pattern of sip changes the return of a point to point trailing return calculation it would do the same for rolling return calculations too.

And just like trailing return it does not help much in comparing two instruments rather than giving some indication about volatility and drawdowns of the individual instruments. Sure you can use it for comparison, averages or not, but the same information can be found from other statistical measures like standrad deviation etc.

 

[nischansr]

@gail76

exact same thing for different date pairs over and over again

And that's exactly why it's so important. That additional set of data points when analysed properly shows more insight into the consistency of performance and the temporality (I use it as a partial indicator to gauge luck vs skill in stock picking).

It makes no sense to compare a single backwards looking xirr value with a single rolling return value for the same period - they are exactly identical. And that's exactly one of the mistakes that people who don't understand how to use rolling returns data do. The other being taking the sequence of rolling returns and then condensing it to a single number via an arithmetic mean and then comparing two funds - that says something about the two funds, but not what they think it does - particularly over longer durations.

Where I find utility in rolling returns is to look at consistency/variability of performance. And when the entire sequence for the same time periods is compared apples to apples between two funds, it does a far better job than xirr in evaluating their relative performances.

Sure you can use it for comparison, averages or not

Avaraging a rolling returns is losing out the one major insight it provides - temporal performance - why would one want to do that?

It's like taking an audio wave for a song and then finding it's average frequency and replacing the whole song with a monotone at that frequency. Its in no way a good representation of the underlying information.

but the same information can be found from other statistical measures like standrad deviation etc

Nope. The temporal behavior is lost when you take a standard deviation and that's valuable information (for me). Sure there are other metrics available which can contain the same info, but they aren't readily available - and xirr, or sip returns don't qualify for that.

Look I'm not trying to convince you of anything. You should use what metrics you like. You are not going to convince me that the rolling returns contains the same information as a std or a single xirr number because it simply isn't true (wrangling with means and standard deviations is part of what I do for a living).