Feeds:
Posts
Comments

Archive for the ‘iPredict Analysis’ Category

Following on from previous posts, another short update on the iPredict stocks for National and Labour to win the 2014 election.

Daily average trade prices for National broke through the 80c barrier on 26 June, and Labour went below 20c on the same day.  The prices were reasonably stable around those levels for about six weeks until 13 August 2014 when the Dirty Politics scandal broke and National took a steep hit.  Average daily prices for National were below 70c for several days, and bottomed out at about 64c on 22 August, well before Collins’ resignation on 30 August.  The stocks have since rebounded, with National today trading for around 84c, an all-time high.

While there was obvious movement, most likely attributable to fear over the fallout from the Dirty Politics scandal, it was short lived.  As mentioned in the previous post the clock is running out for Labour, which needs to find some sort of game-changer, and there is less and less time left before the election for them to do so.

Graph of prices below:

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Since Dirty Politics was released trading volumes and volatility are up significantly. During the last month there have been over 7,100 trades (National and Labour combined), and total volume was over 115,000.  Since opening on 26 October 2011 the total volumes traded are about 195,000 for National and 173,000 for Labour, so almost 1/3 of total volume traded in the last 3 years has been in the last month.  The stocks definitely aren’t moving about on small volumes.

Weekly volume, 2014 National election victory stock on iPredict.

Weekly volume, 2014 National election victory stock on iPredict.

Weekly volatility, 2014 National election victory stock on iPredict.

Weekly volatility, 2014 National election victory stock on iPredict.

The consensus seems to be that this isn’t going to be a particularly close election.

Advertisements

Read Full Post »

Following on from previous posts, another short update on the iPredict stocks for National winning the 2014 election.

Daily average trade prices for National have trended up towards 80c, and Labour down toward 20c.  Prices are as follows:

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

You can see that the prices for National stabalised just above 70c in April and May after earlier peaking at around 76c in mid-March.  I don’t believe that any single event has caused the prices to move.  It’s more a case of running out the clock; Labour needs some sort of game-changer, and there is less and less time left before the election for them to find one.

Weekly trade volumes are up, and now amount to about 1600 shares per week, or $1100 per week.

Daily volume, 2014 National election victory stock on iPredict.

Weekly volume, 2014 National election victory stock on iPredict.

Read Full Post »

Following on from previous posts, another update on the iPredict stocks for National winning the 2014 election.

Daily average trade prices are as follows:

Average daily trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

And zoomed in on the last nine months, starting just before Shearer resigned as leader:

Average daily trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Daily average trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

After topping out at $0.78 on 28 March prices for NATIONAL have stabalised in the low 70% range.

Things have not really gotten better for Labour, with the split now fairly stable at 70/30 odds over the last month.  The book is still a little asymmetrical, stronger on the bid side than the ask side, although there is a bit more room for the price to move around than there was previously.

I still think National’s chances to win are priced a little high, but at 70% it’s close enough that I’m not going to bother throwing money at it to try and fix it.

Read Full Post »

By popular request: an update of the previous post with data from the last eight days.

Average daily trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Average daily trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

And zoomed in on the last seven months beginning on 20 August, just before David Shearer resigned:

Average daily trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Average daily trade price, 2014 election winner stocks on iPredict for National (blue) and Labour (red).

Things have not got better for Labour, with the split now fairly stable at 70/30 odds over the last week.  What is worse is that the book is very asymmetrical: to short National’s chances of victory down to 60% you would need to drop $1813, whereas to bid them up to 80% would take only $553.  Everybody seems particularly bullish on National’s chances.

To be honest I’m a bit surprised.  In the previous post I said that I don’t see how the stock for National can realistically go higher than 70c prior to the pre-election leaders’ debates, and I stand by that.  I’m not sure why people keep bidding the stock up past 70c, although profit taking could be part of it.  I’ve left a few out of the money asks, but am starting to get a bit over-exposed.

Read Full Post »

iPredict is running a contract on National winning the 2014 election.  It was originally launched on 26 October 2011, a month before the 2011 General Election, and has been floating around between $0.40 and $0.60 since then.  It’s only in the last month that the stock has moved significantly beyond $0.60, so it’s worth taking a quick look.  Full trade history is taken from Luke Howison’s excellent API interface for iPredict, and then tweaked with a bit of Excel.

As shown in the graph below, the increase in price since has been pretty constant since it was trading at about $0.45 in October 2013, about a month or so after David Cunliffe was elected leader of the Labour Party.  The average daily price hasn’t dropped below $0.60 since 8 February 2014.

Average daily trade price, 2014 National election victory stock on iPredict.

Average daily trade price, 2014 National election victory stock on iPredict.

In  case you are wondering if this is just a short term fluctuation the answer is probably not.  Daily volumes are shown below.

Daily volume, 2014 National election victory stock on iPredict.

Daily volume, 2014 National election victory stock on iPredict.

Even disregarding the spikes you can still see that trading has increased quite a bit since, again, about mid-October.  In that four month period a total of $19200 changed hands, or about 42% of the $46100 total that has been traded since 2011.  The stock is not just bouncing around on weak liquidity.

It should also be mentioned that bid order book is quite solid: if you wanted to force the evaluation of National’s chances of winning the election back to an even 50/50 then you would need to drop a total of $1420 on 2500 contracts at an average price of $0.57.  Then you’d have to keep it there.  It’s doable, but I don’t fancy your chances.

Intra-day volatility is shown below:

Intra-day volatility, 2014 National election victory stock on iPredict.

Intra-day volatility, 2014 National election victory stock on iPredict.

The most volatile days were 28 January 2014 (David Cunliffe’s State of the Nation speech and “baby bonus”, the new flag debate), 26 October 2011 (contract launch, so meta), 28 November 2011 (first weekday after 2011 General Election), 28 October 2011, and 11 October 2012 (opposition parties’ discussion on “manufacturing jobs crisis” and Labour allegations a video exists of John Key talking to GCSB staff in February about their involvement in the Kim Dotcom case), although most on small volumes.

Daily changes are as follows:

Daily price change, 2014 National election victory stock on iPredict.

Daily price change, 2014 National election victory stock on iPredict.

The biggest absolute daily changes (other than those mentioned above) were 29 April 2012 (general debate over whether David Shearer was the correct choice for Labour Leader (see Dim-Post) and the Banks/Dotcom scandal), 22 April 2012 (a series of bad headlines for National, none of which seemed to have any affect on the polls), 13 October 2012 (“manufacturing jobs crisis”, although interestingly also the day before the MSD/WINZ security leaks exposed by Keith Ng hit the headlines?).

The 28 January 2014 in particular was an interesting day: on the day David Cunliffe was giving his State of the Nation/Baby Bonus speech stocks surged 6.6c from 59.8c to 66.4c on near-record volumes as over $1300 changed hands.  Was this the beginning of the end for Labour’s 2014 election chances?

Finally, this is the point where I’m obliged to point out that if you really think that iPredict has a pro-National bias, and that National actually have a less than a 65% chance of winning the election then you should put your money where your mouth is and go and short some stock.  There is plenty of liquidity there for you to bail out should your views change so you won’t be locked into your position long-term; you can’t use that as an excuse either.

For what it’s worth, I personally I think that barring 4-5 good poll results for Labour the stocks are going to settle at around 65c for the next few months.  I don’t see how they can realistically go higher than 70c prior to the pre-election leaders’ debates.  Having said that, I don’t think one or two good poll results for Labour are going to be enough to move them in the opposite direction either.  Labour’s prospects are looking genuinely shabby compared to four months ago (post Cunliffe election), or even 12 months ago when Shearer was still leader.

Read Full Post »

In a blog post a couple of months ago I discussed the January 2010 Roy Morgan political polling stocks on iPredict. Since then the April stocks have closed, so I’d like to take the opportunity to have a closer look at the results.

As with the January stocks, the data for all trades made on any of the April stocks is available through the history option on the API. Like last time, I have simply copy/pasted all the relevant data from the history function on Luke’s API application into a spreadsheet, and then moved it around a bit to get it into an manageable format. The results are available online in MS Excel format [.xlsx, 752kB]. Please feel free to use or modify the file as you like. The file contains one chart and 13 spreadsheets, please see the previous post for an explanation of the data format.

The first thing that should be pointed out is that at no point did the National stock that eventually closed at $1, NAT.APR10.LOW, trade as the most probable outcome; NAT.APR10.LOW typically traded at 27c ~ 36c below the favoured NAT.APR10.MID and NAT.APR10.HIGH stocks. In fact, the day before closing LOW was trading at just $0.0650, compared to $0.4502 for MID and $0.4174 for HIGH.

This behaviour makes an interesting contrast to many other bundles of stocks on iPredict, for example the bundle on the price of 91 Unleaded petrol, where typically one of the stocks in the bundle will trend towards $1 whilst the rest trend towards $0. Or rephrasing in terms of entropy, the entropy of the bundle of 91 Unleaded petrol stocks will trend towards zero, whilst the entropy of the bundles of Roy Morgan polling stocks will be constrained by a theoretical lower bound. The reason, of course, is that with the petrol stocks the only source of uncertainty is temporal uncertainty, whereas with the polling stocks there is also measurement uncertainty and statistical uncertainty. It is for this reason that the Roy Morgan polling stocks provide such a good laboratory to test the validity of predictions in a futures market.

Once again, the focus of the analysis here will be on the entropy of the political poll bundles over time. For an explanation of entropy in prediction markets please refer to the previous January stocks’ post.

The plot of entropy against time for the April stocks is shown in the first graph below.

Graph showing movement of the total entropy of the 10 political polling stocks

Graph showing movement of the total entropy of the 10 political polling stocks (vertical axis) as a function of date (horizontal axis,) from the opening of the stock on February 19, 2010, up to and including April 8, 2010 - the day before closing. The red series shows the calculated entropy immediately after the last trade of each calendar day (NZT), and the black line shows a linear fit to the data series.

The red series shows the total entropy of the two bundles, and the black line shows a linear fit to the data. In order to help interpret the results the entropies of three reference distributions are also shown in the graph. If the most recent trades of both the National and Labour bundles were $0.03, $0.11, $0.37, $0.37, and $0.12, then the total entropy would take the value of 2.6764, indicated on the graph by the green horizontal line. If the most recent trades were instead $0.02, $0.09, $0.40, $0.40, and $0.09, then the total entropy would take the value of 2.4894, indicated by the yellow horizontal line. Finally, if the most recent trades were $0.01, $0.07, $0.42, $0.42, and $0.08, then the total entropy would take the value of 2.3259, indicated by the purple horizontal line.

So how well did iPredict do? The answer is not so different to that from last time. Firstly, the good news is that on average the total entropy of the ten stocks shows a tendency to decline as the closing date approaches, as indicated by the negative gradient of the black linear fit to the data series. The bad news though is that the slope of the data series itself is not negative-definite, which it should be with only very rare exceptions. We can see a sharp drop and trough from March 15th, followed by a rise and plateau occurring around March 24th; a pattern similar to that seen in the January stocks. Once again it’s not obvious why this pattern is occurring, but it seems to be around the time of release of the penultimate Roy Morgan poll in the series before the stock closes, which was released on March 18. Perhaps some traders are underestimating the statistical error on the results of the penultimate poll, and then assuming that the result of the poll that decides the outcome of the stocks will be approximately the same.

Looking again at the graph, another surprise is the precipitous drop in entropy starting on April 4, 2010, approximately 3 days before the release of the relevant poll. It may just be a coincidence, but the drop looks fairly significant when compared to the daily movements over the proceeding 50 days. A similar pattern was observed two days before closing with the last stocks, and as with last time there were no other polls released around this time that could serve as a reference for the trades, so perhaps it was a trader exercising inside information.

To smooth out some of the day to day fluctuations we have combined the data from the January and April stocks together in the graph below.

Graph showing movement of the average of the total entropy of the 10 political polling stocks

Graph showing movement of the average of the total entropy of the 10 political polling stocks (vertical axis) as a function of days until closing (horizontal axis.) The red series shows the average of the calculated entropy for the JAN10 and APR10 stocks immediately after the last trade of each calendar day (NZT), and the black line shows a linear fit to the data series.

The red series shows the total entropy of the two bundles, and the black line shows a linear fit to the data. The horizontal axis now represents the number of days before the stock closes. The green and yellow horizontal lines are those from the above graph corresponding to entropies of 2.6764 and 2.4894, respectively. The additional two reference entropies are given by the blue (2.9673) and orange (2.8057) lines.

Averaging the two different time periods smooths out some of the fluctuations, but once again we can still see the dip, trough and rise lasting from approximately 30 days before closing until approximately 15 days before closing. This discrepancy indicates an interesting strategy for beginners who want to take part in the stocks: simply wait until three weeks before closing, and if the highest stock in either of the National and Labour bundles is priced over about 45c then sell it and buy the rest of the bundle. Other than the non-negative definite changes in entropy, though, the trading behaviour seems fairly reasonable.

In conclusion, traders at iPredict did a fairly reasonable job with the second set of Roy Morgan Poll polling stocks in the new format, which closed in April, although they did not correctly pick the stock in the National bundle that would eventually close at $1 at any point during the period of trading.

Analogous polling stocks for the first Roy Morgan poll to be released in June have also closed, and stocks for the first poll to be released in August will be closing soon, so I will do an similar analysis on those results and publish it here in the near future.

Read Full Post »

In a blog post a couple of months back I discussed the political polling stocks on iPredict. Now that these stocks have closed I’d like to take the opportunity to have a closer look at some of the results of the trading.

The data for all trades made on any of these stocks is available through the history option on the API, or alternatively can be viewed interactively on the brilliant iPredict API application produced by pipe42 (Luke from Pacific Empire.) For this analysis I simply copy/pasted all the relevant data from the history function on Luke’s application into a spreadsheet, and then moved it around a bit to get it into an manageable format. The results are available online in MS Excel format [.xlsx, 1.3MB]. Please feel free to use or modify the file as you like. The file contains one chart and 13 spreadsheets, visible in the tabs at the lower left of the screen when opened in Excel. Starting from the rightmost tabs the data format is as follows:

  • NAT.VLOW, etc.: The spreadsheets contain the raw data from the API trade history: price, number of stocks traded, value of trade, and the date and time at which the trade was executed.
  • Ordered: This spreadsheet shows the merged data for all ten stocks, and shows date and time of trade, price (for the stock traded only,) and then in the 10 rightmost columns the price of the most recent trade for each of the 10 stocks.
  • Entropy: This spreadsheet shows the date and time of trade, and price of the most recent trade for each of the 10 stocks copied from the “Ordered” tab, the partial entropy for each of the stocks as of their most recent trade, and the total entropy.
  • Daily Summary: This spreadsheet simply shows the data from the “Entropy” tab as of the final trade for each calendar day (NZT.)
  • Chart1: Finally, a time series graph showing the movement of the total entropy as a function of date.

There’s a lot of information there worth picking through, but today I would like to focus predominantly on the question of whether or not the market was able to accurately handle the statistical and other uncertainties inherent in predicting future political polling results.

Firstly though, when scrolling through the raw data on the newly-created spreadsheet I noticed a few interesting visual patterns kept popping up, so I thought I’d share some of these before getting into the serious stuff. The following images are actual screen shots of the spreadsheet, see below each for an interpretation and explanation of what’s happening.

Trading history for iPredict Roy Morgan polling stocks from January 15, 2010

Trading history for iPredict Roy Morgan polling stocks from January 15, 2010, showing a single large-volume trade on a single stock of the 10 stocks in the bundles.

This first example just shows a single large-volume trade, with trade-price information in only one of the ten columns representing the ten respective stocks. The trader simply sold 85 shares of NAT.JAN10.VLOW for a total of $5.05, driving the price from $0.0782 down to $0.0500. None of the other nine stocks were affected in the process.

Trading history for iPredict Roy Morgan polling stocks from January 18, 2010

Trading history for iPredict Roy Morgan polling stocks from January 18, 2010, showing small-volume trades on all 10 stocks.

Here a trader has come along and either bought or sold anywhere from a handful to two dozen shares in each of the ten stocks over a period of six minutes or so, creating an interesting diagonal streak across the spreadsheet. This kind of pattern occurs quite frequently in the data. This incremental trading of the stocks as a complete set fairly closely resembles my own trading strategy, although I haven’t checked whether or not this specific set of trades was my own.

Trading history for iPredict Roy Morgan polling stocks from January 5, 2010

Trading history for iPredict Roy Morgan polling stocks from January 5, 2010, showing a single large-volume trade on one of the stocks (VLOW) in the National bundle, followed by an arbitrage atttempt.

Here we have another pattern similar to that in the first example indicating a single large trade on a single stock. However, on this occasion the trade causes the stock price to make a large movement from $0.2822 to $0.2315, a drop of over 5c. This trade is then followed a minute or so later by small trades on each of the stocks in the bundle, indicating that somebody has performed arbitrage to move the bundle price back to rational levels. The trader would have only made a profit of a maximum of 50c (5c multiplied by 10 stocks maximum in each order book,) and more likely 1c~2c on this arbitrage trade though. This visual pattern is perhaps the most frequently occurring pattern in the data set.

Trading history for iPredict Roy Morgan polling stocks from January 21, 2010

Trading history for iPredict Roy Morgan polling stocks from January 21, 2010, showing large price movements on the release of the relevant Roy Morgan Research poll.

Next we have persistent large trades on all stocks, indicating the release of the relevant poll on January 21. The first person to trade after the release of the poll came in at 14:26:04, and proceeded to make roughly $400 profit over the following six minutes.

Trading history for iPredict Roy Morgan polling stocks from January 21, 2010

Trading history for iPredict Roy Morgan polling stocks from January 21, 2010, showing unusual trading behaviour. The stock price fluctuated up and down several times over a period of a few minutes.

Finally, we have a fairly rare visual pattern of oscillating buy and sell orders over a period of a minute or so. In the example shown above the price rises from $0.1192 to $0.1775, falls back to $0.1264, rises to $0.1775 again, before finally dropping back to $0.1264. I have no idea what could possibly be happening here, although can guess at a few likely explanations: 1) somebody got their buys and sells backwards, and promptly realised and reversed their trades; 2) somebody has figured out a way to leave stop-loss and take-profit orders, which happened to kick in sequentially after the first trade was made; or 3) somebody is making large buy/sell orders to try and rort the market maker and get it cough up a bit of free money. I don’t know if this interesting pattern is visible on any of the other (non-polling) stocks on iPredict, but now I know it exists I think I’ll have to go and have a look for it.

Anyway, back to the task at hand – evaluating the performance of the traders at iPredict, specifically, whether or not they were able to accurately handle the statistical and other uncertainties in the Roy Morgan polling stocks. In a blog post a while back admin at iPredict looked at their data and concluded that the prices of trades on iPredict are a reliable indicator of future probability. The analysis was based on one performed by Google in their analysis of Google’s own internal prediction market from 2005. In their analysis Google used entropy as a measure of the “decisiveness of predictions” over time. I will try to replicate their analysis here for the iPredict stocks.

In the case of prediction markets, entropy can be interpreted as a value which reflects the amount of uncertainty in the stocks; higher degrees of uncertainty correspond to higher entropies. For example, a stock at $1 or $0 should be a sure bet, and has zero entropy, whereas a stock at 50c is a coin flip, and has an entropy of 0.35. The combined entropy for the two bundles of Roy Morgan polling stocks is calculated after each trade of any of the stocks by using the most recent trade for each of the stocks at that particular time. The entropy at the close of each calendar day (NZT) is shown in the graph below (or in the spreadsheet linked above):

Graph showing movement of the total entropy of the 10 political polling stocks

Graph showing movement of the total entropy of the 10 political polling stocks (vertical axis) as a function of date (horizontal axis,) from the opening of the stock on November 5, 2009, up to and including January 20, 2010 -- the day before closing. The red series shows the calculated entropy immediately after the last trade of each calendar day (NZT), and the black line shows a linear fit to the data series.

The red series shows the total entropy of the two bundles, and the black line shows a linear fit to the data. In order to help interpret the results the entropies of three reference distributions are also shown in the graph. If the most recent trades of both the National and Labour bundles were $0.38, $0.21, $0.21, $0.10, and $0.10, then the total entropy would take the value of 2.967, indicated on the graph by the blue horizontal line. If the most recent trades instead were both $0.42, $0.22, $0.22, $0.07, and $0.07, then the total entropy would take the value of 2.806, indicated by the orange horizontal line. Finally, if the most recent trades were both $0.44, $0.23, $0.23, $0.05, and $0.05, then the total entropy would take the value of 2.674, indicated by the green horizontal line.

So how well did iPredict do? The good news is that on average the total entropy of the ten stocks shows a tendency to decline as the closing date approaches, as indicated by the negative gradient of the black linear fit to the data series. The bad news though is that the slope of the data series itself is not negative-definite, which it should be with only very rare exceptions. The initial climb from 2.8 on November 5th to 3.0 on November 25th can probably be justified by a citing a lack of liquidity and/or lack of awareness of the stocks by traders. The sharp drop and trough lasting through to January 5th, followed by the rise and plateau of the following week, however, is not so easy to explain away. Obviously an entropy of around the 2.7~2.8 mark was justified by the unpredictability of the result due to the lack of polling over the Xmas/New Year’s period (no polls were released from December 18, 2009, through to the release of the relevant Roy Morgan poll on January 21, 2010.) But the important point is that this unpredictability itself was predictable; every year there is a five to seven week period around the New Year during which pollsters don’t normally release new polls.

This discrepancy indicates that there is probably still money to be made by somebody with a good knowledge of the workings of the Roy Morgan poll; perhaps an insider at the polling company. How to go about making those trades, however, is a different problem.

Looking again at the graph, another surprise is the precipitous drop in entropy starting on January 19, 2010, approximately 48 hours before the release of the relevant poll. It may just be a coincidence, but the drop looks fairly significant when compared to the daily movements over the proceeding 75 days. There were no other polls released around this time that could serve as a reference for the trades, so perhaps it was a trader exercising inside information.

In conclusion, traders at iPredict did a fairly reasonable job with the first set of Roy Morgan Poll polling stocks in the new format, which closed in January, including correctly picking the stock in each bundle that would eventually close at $1 for most of the duration of trading. As for their handling of the statistical uncertainties in the Roy Morgan polling stocks, however, there appears to still be quite a bit of room for improvement.

Analogous polling stocks for the first Roy Morgan poll to be released in April just closed last Friday, so I will do an similar analysis on those results and publish it here in the near future.

Read Full Post »