Baidu predicts open platform stock forecast test

Start now

March 3, 2015, Tuesday, around 10 o'clock in the morning. Four forecast tasks were created for closing data for several stocks using the Predict API tool:

Predict New-n "stock_1a0001_prevision_20150303"-F. /history/1a0001.txt-k "Stock", "Shanghai Index", "stock index", "market"

Predict New-n "stock_300076_prevision_20150303"-F. /history/300076.txt-k "300076", "gqy video"

Predict New-n "stock_300079_prevision_20150303"-F. /history/300079.txt-k "300079", "digital video"

Predict New-n "stock_300080_prevision_20150303"-F. /history/300080.txt-k "300080", "New Daxin material"

Forecast results

Shanghai 1a0001

2015/03/03 on 3341.79

2015/03/04 on 3350.05

2015/03/05 on 3347.32

2015/03/06 on 3351.79

2015/03/07 on 3357.8

2015/03/08 on 3363.81

2015/03/09 on 3366.52

Gqy Video 300076

2015/03/03 on 17.04

2015/03/04 on 16.88

2015/03/05 on 16.85

2015/03/06 on 16.87

2015/03/07 on 16.79

2015/03/08 on 16.77

2015/03/09 on 16.79

Digital Video 300079

2015/03/03 on 16.81

2015/03/04 on 16.72

2015/03/05 on 16.72

2015/03/06 on 16.9

2015/03/07 on 16.85

2015/03/08 on 16.85

2015/03/09 on 17.01

New Daxin Timber 300080

2015/03/03 on 9.32

2015/03/04 on 9.33

2015/03/05 on 9.33

2015/03/06 on 9.33

2015/03/07 on 9.34

2015/03/08 on 9.34

2015/03/09 on 9.34

Predictions vs. Actual comparisons

Because of the large number of entrances and exits, there is no more observation of the value of a few days. Compare only the closing data for March 3:

Actual value of name prediction value

Shanghai 1a0001 3341.79 3263.05

Gqy Video 300076 17.04 16.34

Digital video 300079 16.81 15.75

New Daxin Lumber 300080 9.32 9.22

Q: Why is it so unreliable?

A: It seems not so easy to make predictions.

"Beer and urine don't get wet together, that's the match."

Key words

Baidu has no reason to put out a completely unreliable thing to let everyone waste time. We've heard stories about the predictions of the 2014 World Cup and the forecast of the peak flow of scenic spots.

Let me give you a two-point reason for this inaccuracy:

1. Time particles spend rough

Stock market sentiment has a short and medium length, with days as particles to predict the minute as particles in the mood of change, may be a little bigger. Baidu forecast in the future to support the week as a particle, I am afraid the results are more accurate.

1. Improper selection of keywords

For the Shanghai Composite, I chose the "stock", "Shanghai Index", "stock index", "big market" these few keywords. What do people do when they are not optimistic about the next day and the future? To Baidu to search the "market", "stock"? Maybe to drink a stuffy bar. "Shot after shot", that's the "keyword" we're looking for. Of course, this data is offline, Baidu does not know, that is, collecting more difficult is not necessary. So what happens on the line, especially on Baidu? Well, "Bank Rate", "balance treasure interest", I agree. What else is there? Well, "Thailand travel", "Self-driving tour", I agree. I think, so unconstrained by the given conditions, should also be Baidu does not have a person to play secretly, and put this toy let everyone together to play one of the reasons.

The weekend question asks: I see your March 7 and number 8th also has the data and some still is changing, but these two days is the weekend, the rest. What's going on?

A: I started with no weekend and holiday data on the historical data, the result is that the error is missing too much data, task submission failed. Then the weekend and holiday stock market closed data are set to 0, did not expect, because February 28 and March 1 is the weekend data of 0, leading to the prediction results all the way down, even negative numbers appear! My conclusion is that the business must be continuous without a large number of irregular breaks and no data in the case of daily granularity. In days the Baidu forecast did not put the business on the weekends as a possible interruption as a hypothesis. For the third time, I postponed the data for the last day before the holiday date, which is generally Friday data copied to Saturday Sunday. Naturally, Thursday Friday Saturday Baidu keyword search volume changes and other factors may lead to the program that the value of the weekend is also changing.

Other

I actually chose the stock this field, really lust, want to be rich think more. It should be more appropriate to choose gold.

Which industry or data is more suitable for Baidu forecast, this is a problem. Baidu has told us scenic spots, traffic, football (this is a bit iffy, as the film box office, reference to the golden age of the failure of the example), epidemics, hotel room order these aspects more accurate. You see, a feature of these businesses with Baidu search volume is very large. Others, follow this principle to discover it. Anyway, you will not have a fortune this year, is not measurable, but Andy Lau may be.

Baidu predicts open platform stock forecast test