## Letters to my Berry #5

A post shared by Karthik S (@skthewimp) on

Your biggest milestone in your fifth month is that you started to eat. Beyond the milk that Amma directly provided you, and the formula milk that we had started you on after the doctor’s advice, the fifth month was when we started giving you what I called as “real food”.

You started with this thing called “ragi cherry” which I personally didn’t like too much – it was made out of a flour made by mixing ragi and other cereals with some nuts, etc. We would make a porridge out of this with some sweet element, and the first time I ate it, I said it tasted like soapnut powder.

Initially you made a fuss eating the ragi cherry, but to my utmost happiness, you seem to be yet another banana lover. After only two or three times of my feeding you bananas, all I had to do was to take your silver bowl and spoon and make mashing noises – and you’d immediately start salivating.

This was also the month where you started implementing Amma’s old company’s slogan “moving forward”. Given the size of your head you had trouble holding it up, but you invented your own way of moving forward while still keeping your head to the ground. I tried without success to draw an animal analogy – sometimes it seemed like you were like an ostrich with its head buried in the sand. Ranga said you were like an Aardvark, moving forward with your head on the ground.

One night I’d left you on the carpet with my house slippers at the other end of the carpet. I hadn’t been gone for a couple of minutes when I saw that you’d somehow traversed the length of the carpet and was about to eat my slippers! Yet another day, we had left you in your bouncer and gone somewhere, and you were trying to slide down. Amma stopped you, but the next time you attempted it, we let you slide. And we were amazed with the poise with which you got down to the carpet, never once worrying us that you would hurt yourself!

This was also the month when you attended your first wedding – your aunt Barbie’s. You were such a centre of attraction during some of the pre-wedding festivities that you were tired and slept through most of the wedding. Halfway through both the wedding ceremony and the reception, we sent you home so you didn’t tire further. So apart from the photos taken at the beginning of each session, you unfortunately don’t appear in any photos!

And of course, the biggest event in your fifth month was that you got named. While you had been named even before you were born, and your official name had been submitted to the municipality when you were a day old, we did a small naming ceremony for you. There, the family priest Nagabhushana Sharma made us give you several names.

So there was the maasa naama (month name) which the priest himself decided. You were “Shachi”. Then there was the nakshatra naama (star name), which we had to come up with on the spot with the given starting letter. The starting letter for you was “Go” and Amma quickly came up with “Goda”, which she later elongated to “Godavari”.

And there was the vyavahara naama (trade name) which was supposed to represent one of your ancestors. The day I first met Amma in 2009, she had told me that she wanted to name her daughter Rukmini, after her grandmother. So there was no doubt about this one.

And then there was the nija naama (real name), which of course had to be Abheri. I had to shout it loud three times, and I did that with my mouth close to your ear. Thankfully you didn’t get startled – suggesting you like your name, and you won’t hate us later in life for it!

This is a monthly series that ordinarily runs on my wife’s blog, but since I wrote it this time (for the first time), I’m putting it here.

Earlier editions:

Letters to my Berry – Month#1

Letters to my Berry – Month#2

Letters to my Berry – Prelogue

Letters to my Berry#4

## When a two-by-two ruins a scatterplot

The BBC has some very good analysis of the Brexit vote (how long back was that?), using voting data at the local authority level, and correlating it with factors such as ethnicity and educational attainment.

In terms of educational attainment, there is a really nice chart, that shows the proportion of voters who voted to leave against the proportion of population in the ward with at least a bachelor’s degree. One look at the graph tells you that the correlation is rather strong:

‘Source: http://www.bbc.com/news/uk-politics-38762034And then there is the two-by-two that is superimposed on this – with regions being marked off in pink and grey. The idea of the two-by-two must have been to illustrate the correlation – to show that education is negatively correlated with the “leave” vote.

But what do we see here? A majority of the points lie in the bottom left pink region, suggesting that wards with lower proportion of graduates were less likely to leave. And this is entirely the wrong message for the graph to send.

The two-by-two would have been useful had the points in the graph been neatly divided into clusters that could be arranged in a grid. Here, though, what the scatter plot shows is a nice negatively correlated linear relationship. And by putting those pink and grey boxes, the illustration is taking attention away from that relationship.

Instead, I’d simply put the scatter plot as it is, and maybe add the line of best fit, to emphasise the negative correlation. If I want to be extra geeky, I might also write down the $R^2$ next to the line, to show the extent of correlation!

## Yet another failed attempt at curbing wedding reception queues

For a long time now I’ve been obsessed about queues at Indian wedding receptions. The process at a reception is simple – you get to the hall, and immediately line up. Once you hit the head of the queue, you get to greet the newly wed couple, give them gifts and get photographed with them. Then you’re shown the way to the dining hall where you have dinner and put exit.

When I started my research into reception queues, my aim had been to save the guests at my own wedding the trouble of lining up for too long. As it happened, I’d failed to spot the bottleneck in time, and hence failed spectacularly. Over a few more weddings that I attended, I cracked the mystery, though – the main bottleneck was in the wedding video.

… Then you hear the click of the photographer’s shutter, and start moving, and the videographer instructs you to stay. For he is taking a “panning shot” across the width of the stage. Some 30 seconds later, the videographer instructs you to move, and the bride and groom ask you to have dinner and show you the way off stage.

The embarrassing bit for the guests, in my opinion, is that having struck a photogenic pose for the photo, they are forced to hold this pose for the duration that the videographer pans. Considering that photogenic poses are seldom comfortable, this is an unpleasant process….

So when my sister-in-law got married a couple of days back, I thought it was time to finally put my research to good use, and save her guests the trouble of standing in line for too long. A couple of hours before the reception on Thursday, I went and had a quiet word with the videographer. I told him about how the panning shot held up queues, and so he should make it quick. After a little discussion, he agreed to use a wider angle for the shot, and cut down the time by half.

Around 8 pm, half an hour after the reception started, the queue wasn’t too long. I was secretly happy that my method was working, but there was the possibility that the short queue was down to low arrival rate rather than high process rate. Fifteen minutes later, the queue had built up through the length of the hall, and would remain so for another half hour. My efforts had come down to nought.

It was when I went up on stage to introduce some of my relatives to the bride (I was the cut-vertex in the network between the married couple and these guests, so my presence was required) that I realised what the problem was. The videographer I’d spoken to had been doing his job, panning quickly, but he wasn’t the only one.

There is always a level of mistrust between the families of the couple at any Indian wedding, and this is mainly down to them not knowing each other well. So there is redundancy built in. Usually, each side brings its own priest. The two halves of the couple collect their gifts separately. And most annoyingly for me, each side arranges for its own photographer and videographer.

So the problem was that while our videographer had been panning quickly as instructed, the videographer engaged by my now brother-in-law-in-law was in no such hurry, and was taking his own time to plan. And since I hadn’t engaged him, it wasn’t possible for me to tell him to hurry up.

And so some guests had to endure a long wait in the queue. If you were one of those, my apologies to you – for I didn’t anticipate the double-videographer problem which would hold up the queue. And my apologies once again to those who had to wait in queue at my wedding as well!

## Dreaming on about machine learning

I don’t know if I’ve written about this before (that might explain how I crossed 2000 blogposts last year – multiple posts about the same thing), but anyway – I’m writing this listening to Aerosmith’s Dream On.

I don’t recall when the first time was that I heard the song, but I somehow decided that it sounded like Led Zeppelin. It was before 2006, so I had no access to services such as Shazam to search effectively. So for a long time I continued to believe it was by Led Zep, and kept going through their archives to locate the song.

And then in 2006, Pandora happened. It became my full time work time listening (bless those offshored offices with fast internet and US proxies). I would seed stations with songs I liked (back then there was no option to directly play songs you liked – you could only seed stations). I discovered plenty of awesome music that way.

And then one day I had put on a Led Zeppelin station and started work. The first song was by Led Zeppelin itself. And then came Dream On. And I figured it was a song by Aerosmith. While I chided myself for not having identified the band correctly, I was happy that I hadn’t been that wrong – given that Pandora uses machine learning on song patterns to identify similar songs, that Dream On had appeared in a LedZep playlist meant that I hadn’t been too far off identifying it with that band.

Ten years on, I’m not sure why I thought Dream On was by Led Zeppelin – I don’t see any similarities any more. But maybe the algorithms know better!

## Who do you subsidise?

One basic rule of pricing is that it is impossible for all buyers to have the same consumer surplus (the difference between what a buyer values the item at and what he paid). This is because each buyer values the item differently, and is thus willing to pay a different price for it. People who value the item more end up having a higher consumer surplus than those who value it less (and are still able to afford it).

Dynamic pricing systems (such as what we commonly see for air travel and hotels) try to price such that such a surplus is the same for all consumers, and equal to zero, but they never reach this ideal. While the variation in consumer surplus under such systems is lower, it is impossible for it to come to zero for all, or even a reasonable share of, customers.

So what effectively happens is that customers with a lower consumer surplus end up subsidising those with a higher consumer surplus. If the former customers didn’t exist, for example, the clearing price would’ve been higher, resulting in a lower consumer surplus for those who currently have a higher consumer surplus.

Sometimes the high surplus customer and the low surplus customer need not be different people – it could be the same person at different times. When I’m pressed for time, for example, my willingness to pay for a taxi is really high, and I’m highly likely to gain a significant consumer surplus by taking a standard taxi or ride-hailing marketplace ride then. At a more leisurely time, travelling on a route with plenty of bus service, I’d be willing to pay less, resulting in a lower consumer surplus. It is important to note, however, that my low surplus journey resulted in a further subsidy to my higher surplus journey.

When it comes to markets with network effects (whether direct, such as telecommunications, or indirect, like any two-sided marketplace), this surplus transfer effect is further exacerbated – not only do low-surplus customers subsidise high-surplus customers by keeping clearing price low, but network effects mean that by becoming customers they also add direct value to the high surplus customers.

So when you are pleasantly surprised to find that Uber is priced low, the low price is partly because of other customers who are paying close to their willingness to pay for the service. When you pay an amount close to the value you place on the service, you are in turn subsidising another customer whose willingness to pay is much higher.

This transfer of consumer surplus can be seen as an instance of bundling, but from the seller’s side. Since a seller cannot discriminate effectively among customers (even with dynamic pricing algorithms such as Uber’s surge pricing), the high-surplus customers come bundled with the low-surplus customers. And from the seller’s perspective, this bundling is optimal (see this post by Chris Dixon on why bundling works, and invert it).

So the reason I thought up this post is that there has been some uncertainty about ride-hailing marketplaces in Bangalore recently. First, drivers went on strike alleging that they weren’t being paid fairly by the marketplaces. Then, a regulator decided to take the rulebook too literally and banned pooled rides. As i write this, a bunch of young women I know are having a party, and it’s likely that they’ll need these ride-hailing services for getting home.

Given late night transport options in Bangalore, and the fact that the city sleeps early, their willingness to pay for a safe ride home will be high. If markets work normally, they’re guaranteed a high consumer surplus. And this will be made possible by someone, somewhere else, who stretched their budget to be able to afford an Uber ride.

Cross-posted at RQ

## On holding stocks

I never understood one thing about investment analyst reports – the “hold” recommendation. This is “between” the “buy” and “sell” recommendations (which are self-explanatory), and it tells an investor to hold on to the stock if he already owns it, but not to buy if he doesn’t.

The problem with this is that the difference between buying and holding a stock is small, especially given the current efficiency of equity markets and consequent low transaction costs. The only difference between holding and not holding a stock is that in the latter case, you spend the transaction cost of buying the stock. That is all. Based on this, it is intriguing that the two have remained distinct analyst recommendations for ages now.

I can think of two possible explanations:

1. One can assume that the investor is fully invested (not holding any cash), and so buying a stock means that he has to sell something else in order to allocate capital to this stock. So in other words, the cost of getting the stock into you portfolio is higher than the trading cost itself – it comes in at the cost of another stock. With these increased transaction costs, it’s possible that it’s not worth buying the stock .

2. Analysts hate to admit it (look at the precision with which they dictate price targets), but there is a wide band of error around their estimates of what price the stock will trade at at some point of time in the future. So the buys are those that are much more likely to be trading up than the holds. So by saying “hold” you are saying “yeah this stock might go up, so I’m not so confident about it so don’t bother buying if you don’t have it already”.

But then there is this school of thought that says that analyst’s buy/hold/sell recommendations do not matter at all, and the value they add is in providing the investor access to the company’s management. Matt Levine has written plenty about this, and you should read his latest stuff on this.

Some five years back, I took a piece of advice from Dilbert creator Scott Adams. A few years earlier, he had blogged that there are two ways in which one can be successful in a career –

But if you want something extraordinary, you have two paths:

1. Become the best at one specific thing.
2. Become very good (top 25%) at two or more things.

The post had made an immediate impression on me when I had read it back in 2007. And when I was planning to leave a full-time corporate career in 2011, it was Adams’ old advice that I turned to.

There were a number of things that I’d found myself to be good at (definitely top 25%) – mathematical modelling, data analysis, writing (based on this blog), economic reasoning, financial markets and maybe even programming (I’m a good coder but lousy software engineer). Combining these, I reasoned, I could do very well for myself.

And over the last five years I have done reasonably well for myself. I’ve built a fairly good freelance consulting practice which brings together my skills in mathematical modelling, data analysis and economic reasoning. The same skills, along with an interest in public policy, have led to me joining a think tank as a Resident Quant. Data analysis and writing together has got me a column in Mint. Yet another subset led me to become Adjunct Faculty at IIM Bangalore. And yet another led to my book, which is currently under publication.

However, now that I’ve decided I’ve achieved enough in my portfolio life, and am looking for a full time job (it was supposed to happen a while back I know, but I postponed it due to an impending location change – I’m moving to London in March), I’m not sure this strategy (of being reasonably good in multiple things rather than the best at one thing) is particularly optimal.

The problem is that the job market hasn’t evolved to sufficiently demand people who are good at several things (rather than at one thing). This is a consequence of not enough people following Adams’s second advice – they’ve chosen to strive to be the best at one thing instead.

And so, if you are like me, and consider yourself reasonably good at several things rather than the best at one thing, the job market doesn’t serve you well. Think of all the things you’re good at as dimensions, and your skillset being represented by a vector across all these dimensions. Traditional job markets tend to look at you from the point of view of one of these dimensions (the skill they’re hiring for). And so, rather than showing your potential employer your full magnitude, you end up only showing the projection of your vector along the dimension you’re optimising for.

And if you are good at several things, it means that the magnitude of the vector along any one skill is far smaller than the magnitude of your full vector. And the job market is likely to leave you frustrated!

In contract bridge, when you are dealt a hand that is equally strong in all suits, you bid to play a No Trump game. In this scenario, though, it seems like it’s impossible to effectively play No Trump.