Taiwanese fuel subsidy and the Global Price API

I might have gone a bit overboard with this. Again. It all started when I was reading how the Taiwanese government is planning to implement a fuel subsidy. It bothered me because:

  1. I’m not a big fan of subsidies, since then people cannot make decisions based on the real cost of things
  2. Subsidy comes out of someone’s pocket anyway, so ultimately everyone does pay for it
  3. Looking at the governing party‘s track record and the nearing elections (<1 year), this is likely to be politically motivated
  4. Fuel is pretty cheap compared to other places already, so it must be already heavily subsidized.

The points 1, 2 and 3 are opinions and generalizations. I had to realize that point 4 I just thrown in there without knowing whether it is really true – I just believed it is. Now here’s a good example for [Citation needed].

XKCD: wikipedia protester
XKCD: Wikipedia protester

Filling in the blanks

Yeah, what do I really know? I lived in Hungary and Britain before, and I remember there fuel was more expensive then in Taiwan. Also, it’s open knowledge that the US is pretty cheap, compared to most of the world, that is… But are there any patterns in the price, and what would I expect Taiwan’s level to be?

I was checking around for a while, fishing for the right keywords for the search (“fuel” / “petrol” / “gas”, depending where one’s from, as a starter). Found a couple of sites but they were mostly looking at price comparisons within a country (like Petrolprices for UK) or within a region (like AMZS for Europe, works weird – no real static link, click the UK flag then “fuel prices” in the menu on the left). After a while, however, I did come across a German organization, Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) which in fact had a study about international fuel prices. Go ahead and check the 2009 version, it is very interesting to see all that data and some analysis as well. Historical trends, relative prices, some case studies too. I would have loved to see more analysis, but maybe next time, when I’ll be in the in a position to commission such research. :)

Anyway, looking at the Taiwanese data in from 2009, diesel is slightly subsidized while gasoline is, which means that they are somewhat – but not too much – below and above the US price, respectively. In the ranking, Taiwan is very much on the cheap end.

While looking around a bit more on their site, they have a Data preview for 2010/2011 as well. The more recent is the data the better. Took a look at that too. Now Taiwan is a bit above of the US price with both types of fuel, but still on the cheap end. All their data was in a picture, though, that’s not very handy… So did a little data entry, resulting in this datafile.

I was also thinking, whether the economy affects the prices, and if does then how? Wikipedia first for GDP/capita for all the countries of the world, but they were taking the data from the International Monetary Fund, so let’s go to the source. I checked out their data export tool and there were quite a few more fields to choose from. I went with GDP per capita, Implied purchasing power parity, Value oil exports, Value of oil imports (these last two are good catch:). The output is here. The bottom of the page has a download link, downloaded it into this file without modifications.

Next, had to write some analysis code for the whole thing as well, converting the data into suitable database, fixing some errors in the primary database’s data formats. So in the end I had a simple little script that does:

  1. Clean up some of the names (some Unicode errors originally), and fixes one: I do want to convert the original “Taiwan Province of China” into “Taiwan”…
  2. Fix formatting errors in the IMF  data: they used number formats of “12,345.00” instead of just “12345.00”
  3. Fill missing values with “-1” so it’s easy to filter out later
  4. Rank countries by fuel price, ignore countries that have missing economic data or missing price data
  5. Combine all this and print out on the console

(Scripts and data are shared in this git repo.)

Taking a look

I was looking around for some useful visualizer – something that can handle this much data better then an ordinary static plot. Fortunately, Highcharts JS seems to just the right thing…

The first plot I wanted to reproduce is the one from GIZ’s Data preview. Let’s see how it works out:

Instructions: “red”/”blue” countries are net oil exporter/importer respectively, hover over any of the lines to see which country it is, can click-and-drag zoom into area…

So yeah, Taiwan is down at 49 out of 161 countries, and just few net-importer (blue) countries ranking higher. Even those are mostly poor ones.

Now the second picture, how do fuel prices compare with GDP/capita – which I naively think to have some connection to economical power:

Instructions: note that the GDP scale is logarithmic, hover and zoom are the same as before.

Might be just my eyes, but it seems to me that there are two lines on this plot if one ignores the net exporter (red) countries for a second. From the middle to the right prices are increasing: the wealthier countries can pay more for the fuel. On the other hand, from the middle to the left prices are increasing again: poorer countries cannot really afford it. The cheapest (nominally) are the middle-to-poor, $1000-5000 GDP/person countries.

Bottom line: Taiwan is there at ~$18000/¢100, and if there are indeed these lines, then Taiwan is waaaay below the wealthy country line. Based on the economy, the price should be closer to ¢150. This suggest to me that the original assessment was correct: Taiwanese fuel is cheap.

Global Price API

All of this data-hunting and conversion and plotting should not be this much of a pain. I have a feeling there’s need and desire for open access for such information and that transparency would help people’s decision making – whether those people are in charge or part of the public (and should be “ultimately in charge”). Of course, I’m not the only one to say this, and I’m not even a very good one making this happen – just check out Hans Rosling’s TED talk.

I was thinking, how to build a globally accessible database of consumer prices? Fuel is a good choice because it’s universally needed and there are not too many kinds, one can compare apples to apples. On the other hand, there could be other items as well. Maybe recruit a few volunteers from a big bunch of countries so periodically they add more info to a database. Or fully crowdsource it, maybe even the item categories as well? Then build an interface that it can be easily queried and used by programmers and non-programmers as well. Or is there any such database already? Pitching version: “Archive.org for global price and other public data”. Not that I have a business model for this… I’m sure I’d prefer the same as SimpleGeo and completely open the data itself, but I know there are people who will still find opportunities – or make some.

Any thoughts on this? One selfish thought I have is that this would be lovely so I never ever again have to manually enter all the names of all the countries of the world. :) But I do believe there’s much more to this project.

Hacker learning Chinese

I guess when a hacker learns a language it is different from the way “others” do that. I guess this, because I think I’m a hacker, I’m learning a language and it feels different. I see two main factors coming in the picture:

  1. I’m connecting things in my life, so the things I do usually need some motivation or purpose behind them, without which they are abandoned. The activities I keep up the best are the ones that connect multiple different things.
  2. I want to do things as efficiently as possible. Hack the tools, hack the process to make it better. If there are no tools, make some. If no processes, come up with ideas.

Okay, these are pretty vague expressed like this, now let’s give the example that prompted this post: I’m learning Chinese. Living in Taiwan for two and a half years now, so “high time” doesn’t even start to describe it. Finally I got a tutor, and a good one at that, so twice each week I have a good session of chatting and learning. After each of the sessions we have at least 30 or 40 new words and expressions written down. Those would normally be just forgotten, so I take an effort (about another hour or so after the session) to type them into a Google Document, this very one on the picture:

Google Docs Chinese vocabulary
The Google Document that powers my learning (click to enlarge)

Enter the English expression, the Chinese original and the pronunciation in Bopomofo (which I prefer to Pinyin). This last is possible because of Yahoo Chinese-English Dictionary (one service that is generally better than Google’s own Translate, though I frequently need to use both). The other three columns I’ll came in just recently.

At the moment I’m up to 307 expressions, and that’s just not possible to practice from a spreadsheet like that. I remembered, though, checking out a fellow StartupBus participant, Pamela Fox‘s Google Spreadsheet Flash Cards some time ago. It was fun but followed a different logic than me so couldn’t make complete use of it. But then: why not make a new practice system for myself? This goes back to the precious points: 1) connecting programming and languages – in both I’ll learn something new and they’ll reinforce each other’s motivation, 2) use the exact tools that I need even if I have to make them (because it’s possible to do).

Also after having done Who Said That? (that is currently down due to the AWS fail), I was into guessing games: let’s make a vocabulary guessing practice app that uses the above spreadsheet that I have anyways. What format it should be? Well, for the very first test, to see how to interact with Google Docs and such, just made it as a simple, console-based python app, something like this:

Chinese learning console
The console app to practice me Chinese vocabulary (click to enlarge)

Each practice round has 30 questions, randomized, 4 possible solutions, with the pronunciation as hint. Simple, though very ugly in the inside at the moment (here’s the repo, I still need to write a ReadMe). It works and now I know what to look out for in terms of implementation (how to log in, how to get and update data, stuff like that).

The other 3 columns in the above spreadsheet are explained as well: they are keeping score such as the total number of times a word/expression is practiced, the number of good answers and the number of current good-answers-in-a-row. These provide a rough-and-ready way to diagnose and manage the learning process – until I come up with better ways. Anyway, right now I get about 50%-70% good which is more than I expected, but still a lot more room for improvement.

A list of improvements to the program that I’m thinking about:

  • Making it into a site, so I can use my phone on the go to practice. Also, potentially others can use it as well to practice anything based on any spreadsheet with “one side – other side – hint” structure.
  • I’m pretty sure this whole things could be done in a single Javascript powered page. I don’t know enough Javascript to pull it off yet, but I’ve seen that all the components separately, and that would make a very portable and compact solution.
  • Need to figure out some easier setup of the spreadsheet if this is to be used by others later. I cannot rely on them understanding what they way I was thinking. Maybe in-app option to add more fields?
  • I remember reading somewhere that the most effective practice is that I reduce the frequency of checking words that I know well. That’s where the last column comes in: as one has more right answers in a row, one can tast that word/expression fewer times. If there’s a mistake then go back to the original method and test it more until score builds up again. This can potentially be a very complicated algorithm, I got to think of a way that scales well (ie. it is not too bad compared to an ideal method but does not require extensive amount of calculation). Have some ideas, but they need more polish.
  • After watching Salman Khan’s TED talk this year it grabbed me just how much information is there in one’s actions (they do amazing feedback to teachers on how the students study), how much better you can understand why did people what they did if you have all those diagnostic information (ah, the temptation of Big Data:). To apply this idea of extended diagnostics I could have a logging system instead of keeping (a simple) score. From there the system could get: how much you practice, how are you doing / getting better, what are easier or more difficult words for you, what two words you mistake with one another, suggest things to practice more, suggest words from topics that you know well to extend your knowledge… And more (this was just a little brainstorming). I don’t think I’d have time to extend it like that, and ideas are a dime a dozen, but one never knows…
  • Adding more modes of learning not just multiple choice, multiplayer learning, more game mechanics (achievements, pins anyone? :)
  • If there are central datasets instead of self-provided ones, then the system can anticipate what are the difficult parts from other students’ performance before you.

Now, let’s just see how will my Chinese improve during all this hacking, since that’s the main point, isn’t it? :)

New Formosa Restaurant Signature Dishes
Some motivation for learning, loads of Taiwanese food :)

Ps: If you have any language learning tricks, let me know in the comments!