[討論] 未來事件期貨交易
想看原文的板友,原文在下半部。上半部是重點敘述跟感想。若是
我有誤讀或敘述不完善的部份,還請提出指正。
原本我想貼原文再條列一兩個重點就好,可是先前中華隊跟中信鯨
隊打熱身賽中華隊打者狀況比想像中的還要低迷,竟然只以2:1獲勝
,害我輸掉近兩萬P幣。為了賺P幣繼續賭博,我只好多打一點字了。
部份媒體對未來事件期貨交易站intrade(www.intrade.com)讚譽有
加,因為intrade的有過顯著的預測成果。最讓人印象深刻的例子是
在2004年的大選,intrade認為布希有50%以上機會能夠獲勝的州,布
希全部拿下,勝選機會被認為不到50%的州最後也都由凱瑞拿下,預測
命中率高達百分之百。到2006年期中大選時,當各個政論結目來賓認為
共和黨能夠守住參院時,intrade的玩家則已經預測到最後民主黨的勝利
了。看起來群眾的集體智慧比少數專家還精準。
(內文批評到預測市場對2006年參院結果預測失誤,那是指另一個平台
Tradesports)
不過intrade也是有過極為明顯的失誤,如今年2/15加州初選,出口
民調釋出Clinton女士以3%左右的幅度領先時,intrade的玩家認為Obama
有60%的機會贏得加州。更丟臉的失誤則是New Hampshire,Clinton女士
表現超出民調的預期,以2%的差距逆轉獲勝;但在結果揭曉前,intrade
的玩家竟認為Obama有99%的機會贏得該州初選。
本文認為intrade會出現這樣的誤差原因有兩個,首先是獎勵誘因不足,
再來是,樣本不足,且意見來源的同質性過高。
intrade上的交易量比起股市動輒數兆的金錢流量比起來明顯過低,不少
交易金額都只是上千美元的規模。因此人們不會非常積極的吸收資訊並
在投資市場上做反應。Intrade市場常出現的現象是股價不是對市場反應
過慢就是過度反應。最明顯的例子是Obama在Wisconsin獲得出乎意料之外
的大勝,intrade上認為Obama最後會贏得提名的機率維持在65、70的水準
好幾天後才上漲到目前的85。如果牽涉的金額夠大,投資者不會放掉Obama
贏得Wisconsin初選這個重大轉捩點,趁Obama股還便宜時即早買進來大
賺一筆。過度反應的例子,前面提到的玩家認為Obama有99%的機會拿下
New Hampshire非常明顯。
另一個同時過度反應跟反應過慢的例子是intrade對Rhode Island民主黨初選
的預測。這項預測對19號Wisconsin的結果反應快速,認為Obama 在該州勝選
機率上漲到80%。然而玩家卻沒考慮到Rhode Island位在Clinton女士強勢
的東北角,選民結構上有不少藍領勞工,Clinton在那裡有一定的優勢,20號
左右的市場行情顯然是過度反應。然而前兩天新聞指出Rhode Island有相當
數量教育程度高、無黨派立場的新選民打算參與初選投票,新民調也顯示
Clinton女士在該處只領先9%,Obama的行情跌到22.5%不僅是過度反應,
在這些新資訊釋出時行情也不見上漲,顯然反應又過慢了。
市場規模小更大的問題是,特定候選人的狂熱支持者用相對少量的資金
就可以讓整個預測市場失真。比方說共和黨參選人Ron Paul雖然民調支
持率不高,可是在網路上卻有不少狂熱支持者喊著媒體跟民調對Paul有
偏見,Paul最後必定會贏得共和黨提名。intrade上Ron Paul被預
測會贏得提名的機率也在這些狂熱者的影響一度高達10%。然而任何了解政治
現實的人都明白Paul不管在哪一個時間點贏得提名的機率都遠遠小於10%。
只要市場夠大,類似的狂熱支持者要扭曲市場便需要更多的資金,預測被
扭曲的機會便將大幅降低。
如果intrade繼續成長,牽涉到的資金流量更大,對事件的反應也會更快速
更真實。目前intrade的規模已比2004年來得大,光是這兩個月的初選進行
下來,牽涉到的資金已比2004年來的多。
再來是意見來源的同質性過高,本文認為預測市場要成功,參與預測的
玩家除了整個樣本要夠大,另外投資玩家跟能夠決定事件結果的參與者
在組成結構越接近,預測也會更形準確。比方說麥可傑克森的性騷擾案,
投資玩家的結構跟能決定判決結果的12位培審團成員肯定有很大的不同,
因此期貨市場上的賠率要精準的反應判決結果是不可能的。
參與期貨市場的玩家多半是有錢有閒,知識水準相對較高的族群。當這些
背景相近,數量不多的玩家在對各州的選情做預測時,這些很有可能不是
當地人的投資者很難根據地方上輿論或候選人動員情況的掌握來做預測;
只能根據同樣的報導、同樣的民調、同樣的部落格資訊來做判斷。當
報導、部落格、民調對選情判斷錯誤時,預測市場也只能跟著犯錯。
不過至少對於選情的預測,專家學者除非神通廣大,知道某候選人陣營極
將放震撼選情的大決(話說回來,知道內情的人也通常不會上媒體透露這
些內線訊息),專家學者也只能憑據著同樣的報導、部落格、跟民調做判斷
。既然判斷的依據相同,拿真錢下去賭的多數玩家總比少數有立場的專家
可靠點。
至於intrade上類似哈馬斯什麼時候會承認以色列,什麼時候以色列、巴基
斯坦會搞和平談判,這些項目都不必太認真對待;除非玩這兩項交易的
都是哈馬斯跟以色列方面的高級決策者。
目前台灣有政治大學跟中央研究院在做類似的預場市場,顯示學界對研究
預測市場對政治事件的預估也相當有興趣。雖然引進的相關專家很熱血
的上媒體表示,預測市場對選情判斷的準確度比傳統民調高上75%;不過
很遺憾的類似的東西移殖過來恐怕無法同樣神準。
除了參與人數少以外,最關鍵的是或許由於法令限制,這兩家學術機構
玩得是虛擬貨幣,而非玩真錢。雖然在註冊帳號限制上或許可以限制
一人一帳號,不過若是特定陣營要影響預測結果,只要動員群眾上來就
可以輕易扭曲,反正是玩虛擬貨幣,根本不痛不養。
台灣學界如果要辦好預測市場,沒克服不能用真錢的問題是很難的。
眼下最好的解決方式我建議在ptt辦,用p幣玩。p幣對ptt鄉民的份
量肯定還遠比兩家學術單位預測市場的參與者對該處用的虛擬貨幣還要
重很多。ptt的鄉民數量已比去年同時間大幅成長,十幾萬的鄉民以預
測市場的樣本來講不算小了,雖然可能背景都是大專生,分歧度不足;
再加上p幣要靠打字來賺,像我這樣的虛擬賭博狂熱者也會
因為錯押中華隊p幣破產在螢幕前整整哭了10分鐘;同時ptt的博羿市場也
頗為活躍,預測市場開在這裡有興趣的鄉民一定也不少;更理想的是p幣雖然
不是真錢,不具交易價值,鄉民仍然對p幣相當熱情,這些條件綜合起來,台
灣學界要辦好預測市場ptt是最後的希望。
===
對台灣學界引進預測市場的討論,我已經試著盡量不牽涉到現實政治
,如果板主認為不妥,我會修文刪掉這部份。
Why Prediction Markets Fail
http://bigpicture.typepad.com/comments/2008/01/prediction-mark.html
Friday, January 11, 2008 | 07:50 AM
in Markets | Politics | Psychology/Sentiment
Over the years, I have been critical of prediction and futures markets. In
particular, the specific ways certain parties misuse them (i.e., politics).
However, I am a big believer that markets can generate valuable economic and
investing data that can be quite helpful when handled appropriately. In my
own work, we rely on the 10 year interest rate as part of our overall
economic modeling. On the equity side, we use various trend component for
indices in addition to volume, money flow, short interest and institutional
ownership for sectors and specific companies.
In each of these cases, we are relying on markets that are deep, diverse and
trading in dollar volumes measured in trillions. This is part of my criticism
about future markets: They are thin, trading volumes are anemic, the dollar
amounts at risk are pitifully small. Thus, these markets are subject to
failure at times.
Indeed, the excuse making for the failure of the futures market began almost
as soon as the New Hampshire primary ended (see Why the Online Prediction
Markets Blew New Hampshire, Prediction markets are forecasting tools of
convenience that feed on advanced indicators, etc.)
None of these really looked at what makes futures markets work, how they can
fail, and what their strengths and weaknesses are. This morning, I want to
delve into some the problems these markets can have, why they fail, and what
value they do offer.
First, Let's start by reviewing some of the more spectacular prediction
market failures:
‧ Iowa Primary 2004 (Howard Dean vs everyone else)
http://bigpicture.typepad.com/comments/2004/01/iowa_and_predec.html
‧ 2006 GOP Senate Retention
http://bigpicture.typepad.com/comments/2006/10/odds_of_gop_ret.html
‧ New Hampshire Primary 2008 (Obama vs Hillary)
http://bigpicture.typepad.com/comments/2008/01/confusing-cause.html
In the area of politics, I am in agreement with Dan Gross, who wrote "these
are less futures markets than immediate-past markets." In other words,
bettors essentially aggregate polling data that is already out -- rather than
forecasting the unknown future.
When it comes to politics, its not as if the Iowa electronic market or
Intrade gamblers have any special insight or inside knowledge. They don't
"know" anything -- as individuals or collectively -- that the rest of the
public (insiders included) don't already know. They're all reading the same
newspapers, blogs, polls, etc., and responding to whatever broad narrative
happens to be coming out that day or week. They respond just as any other
focus group would, off of the same information the electorate has.
I sometimes think of the political futures markets as a focus group unto
themselves. Here's where things get really interesting: When the group is
something less representative of the target market, they get it wrong with
alarming frequency. Indeed, the closer the traders are as a group to the
target decision makers/voters, the better their track record.
Consider smaller "electorates" beyond the political events mentioned above.
Think of the Michael Jackson Trial, or the Morgan Stanley CEO Purcell
resignation betting -- these much smaller groups of 12 jurors, or a handful
of MS directors don't lend themselves very well to focus groups or polling.
Hence, the lower success rate.
Back to political futures: I suspect the traders at Intrade in 2006 missed
the GOP Senate loss, because as a group, they themselves skewed away from the
rest of the nation. The demographics of Intrade traders are likely higher
income, better educated, leaning more GOP than the USA as a whole. Hence, the
variation of this focus group away from the larger electorate led to the bad
forecasts there.
We see this especially in close elections: When they are similar to the
voting electorate, they get it right; as they differ, their error rate goes
up. This commonality to where they fail is one that's worth exploring further
by some suitable academic (Robin? Justin?)
A more academic way to analyze these issues is to look at what makes markets
work. Lets go to Legg Mason's Michael Mauboussin:
"The "wisdom of crowds" is a colloquial way of describing the market as a
complex system. The work on wisdom of crowds shows that when certain
conditions are met—diversity, aggregation, and incentives — markets tend to
be efficient. Conversely, when one or more of those conditions are violated,
markets can and do become inefficient (i.e., price is no longer an unbiased
reflection of value).
For a host of social psychological and sociological reasons, diversity is the
most likely condition to be violated. Here, too, market observers have
recognized the importance of diversity. But what’s essential to recognize is
the relationship between diversity breakdowns and asset price performance is
nonlinear. Diversity can be on the decline as the asset price rises, which
makes the market fragile. Like a rubber band that’s stretched, the tension
builds and the inevitable snapback is often painful."
Hence, we see more potential sources of market failure as these conditions
are violated:
1) An insufficient amount of incentives;
2) A lack of diversity of ideas;
We've discussed both of these, albeit in less than academic terms: The small
thinly traded markets means the money at risk is rather small -- measured in
$1000s, not trillions of dollars. Hence, the incentive is minimal -- or
perhaps non-monetary. Maybe this attracts a somewhat more politically
passionate -- and less objective -- group of traders.
That can lead to a lack of intellectual diversity. As Mauboussin discussed,
that is the most likely source of market failure. And as we review above,
this is particularly acute when it comes to smaller groups not paralleling
the demographics and or politics of Futures markets' traders.
~~~
In some ways, Futures markets are similar to many Wall Street economists: at
times, they merely extrapolate the current trend forward. Hence, they can be
right for most of a run -- an economic expansion or stock bull market, etc.
However, this process leads them to completely miss major turning point, or
an unexpected shift or event. I suspect this factor also played into the New
Hampshire 2008 failure...
--
■所有荷蘭人如果每週一天不吃肉,就可達到荷蘭政府希望家家戶戶一年所減少的二氧化
碳排放量目標。
■南美洲約有四億公頃的黃豆作物是種給牛吃的;如果是提供給人類食用,則只需兩千五
百萬公頃就可以滿足全世界所需。
「不吃肉、騎腳踏車、少消費,就可協助遏止全球暖化。」 by Dr. Rajendra Pachauri
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 122.127.66.154
※ 編輯: swallow73 來自: 122.127.66.154 (03/02 13:03)
※ 編輯: swallow73 來自: 122.127.66.154 (03/02 13:29)
推
03/02 13:09, , 1F
03/02 13:09, 1F
推
03/02 13:14, , 2F
03/02 13:14, 2F
→
03/02 13:16, , 3F
03/02 13:16, 3F
→
03/02 13:17, , 4F
03/02 13:17, 4F
→
03/02 13:17, , 5F
03/02 13:17, 5F
→
03/02 13:18, , 6F
03/02 13:18, 6F
→
03/02 13:18, , 7F
03/02 13:18, 7F
→
03/02 13:20, , 8F
03/02 13:20, 8F
※ oodh:轉錄至看板 politics 03/02 13:20
→
03/02 13:29, , 9F
03/02 13:29, 9F
→
03/02 13:30, , 10F
03/02 13:30, 10F
※ 編輯: swallow73 來自: 122.127.66.154 (03/02 13:42)
推
03/02 13:39, , 11F
03/02 13:39, 11F
推
03/02 14:08, , 12F
03/02 14:08, 12F
→
03/02 14:09, , 13F
03/02 14:09, 13F
→
03/02 14:26, , 14F
03/02 14:26, 14F
→
03/02 14:26, , 15F
03/02 14:26, 15F
→
03/02 14:27, , 16F
03/02 14:27, 16F
→
03/02 14:28, , 17F
03/02 14:28, 17F
推
03/02 14:28, , 18F
03/02 14:28, 18F
※ 編輯: swallow73 來自: 122.127.66.154 (03/02 14:33)
推
03/02 15:15, , 19F
03/02 15:15, 19F
→
03/02 15:16, , 20F
03/02 15:16, 20F
推
03/02 15:17, , 21F
03/02 15:17, 21F
→
03/02 15:18, , 22F
03/02 15:18, 22F
→
03/02 15:18, , 23F
03/02 15:18, 23F
→
03/02 15:20, , 24F
03/02 15:20, 24F
→
03/02 15:22, , 25F
03/02 15:22, 25F
→
03/02 15:45, , 26F
03/02 15:45, 26F
→
03/02 15:45, , 27F
03/02 15:45, 27F
→
03/02 15:46, , 28F
03/02 15:46, 28F
推
03/02 15:47, , 29F
03/02 15:47, 29F
→
03/02 15:48, , 30F
03/02 15:48, 30F
→
03/02 15:49, , 31F
03/02 15:49, 31F
→
03/02 15:51, , 32F
03/02 15:51, 32F
→
03/02 15:53, , 33F
03/02 15:53, 33F
→
03/02 15:55, , 34F
03/02 15:55, 34F
→
03/02 15:56, , 35F
03/02 15:56, 35F
→
03/02 15:57, , 36F
03/02 15:57, 36F
→
03/02 15:57, , 37F
03/02 15:57, 37F
→
03/02 15:58, , 38F
03/02 15:58, 38F
→
03/02 15:59, , 39F
03/02 15:59, 39F
→
03/02 16:00, , 40F
03/02 16:00, 40F
→
03/02 16:01, , 41F
03/02 16:01, 41F
推
03/02 16:03, , 42F
03/02 16:03, 42F
→
03/02 16:02, , 43F
03/02 16:02, 43F
→
03/02 16:03, , 44F
03/02 16:03, 44F
→
03/02 16:04, , 45F
03/02 16:04, 45F
→
03/02 16:05, , 46F
03/02 16:05, 46F
→
03/02 16:05, , 47F
03/02 16:05, 47F
→
03/02 16:06, , 48F
03/02 16:06, 48F
→
03/02 16:07, , 49F
03/02 16:07, 49F
→
03/02 16:07, , 50F
03/02 16:07, 50F
→
03/02 16:27, , 51F
03/02 16:27, 51F
→
03/02 16:28, , 52F
03/02 16:28, 52F
→
03/02 16:28, , 53F
03/02 16:28, 53F
→
03/02 16:36, , 54F
03/02 16:36, 54F
→
03/02 16:37, , 55F
03/02 16:37, 55F
→
03/02 16:38, , 56F
03/02 16:38, 56F
→
03/02 16:38, , 57F
03/02 16:38, 57F
→
03/02 16:39, , 58F
03/02 16:39, 58F
→
03/02 16:41, , 59F
03/02 16:41, 59F
→
03/02 16:42, , 60F
03/02 16:42, 60F
→
03/02 16:42, , 61F
03/02 16:42, 61F
→
03/02 16:45, , 62F
03/02 16:45, 62F
→
03/02 16:45, , 63F
03/02 16:45, 63F
→
03/02 16:46, , 64F
03/02 16:46, 64F
→
03/02 16:46, , 65F
03/02 16:46, 65F
→
03/02 16:47, , 66F
03/02 16:47, 66F
→
03/02 16:47, , 67F
03/02 16:47, 67F
推
03/03 20:37, , 68F
03/03 20:37, 68F
→
03/04 03:37, , 69F
03/04 03:37, 69F
推
03/04 21:25, , 70F
03/04 21:25, 70F
→
03/09 09:05, , 71F
03/09 09:05, 71F
→
03/13 00:20, , 72F
03/13 00:20, 72F
討論串 (同標題文章)