Topic: : - page 23. (Read 70881 times)
The zero-fee bots still seem to be going strong. If I don't see a change in spreads over the next few days I am going to have to presume we can expect another month of this. This will weigh heavily in my analysis going forward.
I'm torn about this plateau. While I like the fact that we can point to the price history and tell people "Look, stop waving your hands, we've been here over a month without massive swings." I'd also like to see some kind of trending.
Can't have it both ways, I guess. I assume we're basing like we did before the upswing through parity.
It doesn't do that well when there is no data. A flat line does not contain very much data.
I think I formulated my reply poorly, what I meant was that just because your model is not doing well on a short period it does not mean it won't long-term. On some specific intervals the model should perform worse than doing nothing.
Well, I'll tell you what then. There is one thing that has worked better than anything I have come up with for predicting the day's price, at least lately... That is, yesterday's price.
Then your models are no good..?
Right now the daily model does not produce an output that is more predictive than the previous day's price.
Even (the price of yesterday)*0.25 + (price of today)*0.75 should outperform that model?
I don't know how you evaluate so the problem might lie there? Of course on some specific days your model will perform worse than doing nothing but if it is doing so over time then it is not good at all...
attempts at price manipulation...
Does today's activity not look like normal market activity?
at the front end I would expect a variance of no more than 5% to 10%, but a the back end more like 10% to 20%...
2011-07-23 $13.386 $14.096 actual: 13.51 13.77
2011-07-24 $13.208 $14.331 actual: 13.6 14.1
2011-07-25 $12.903 $13.995
2011-07-26 $12.778 $13.559
2011-07-23 $13.386 $14.096 actual: 13.51 13.77
2011-07-24 $13.208 $14.331 actual: 13.6 14.1
2011-07-25 $12.903 $13.995
2011-07-26 $12.778 $13.559
Maestro!
If the trend is precisely within your modeled range, as it has been, does this new conforming data add very much information/precision going forward?
So how does your model respond to the spikes today?
So far it is no factor, because I am looking at TradeHill numbers for my analysis, and this has not yet resulted in a confirmation on TradeHill. The way I think about it, attempts at price manipulation are not really successful unless the changes they produce propagate to the other exchanges. If noise from those attempts and noise from other exchange 'errors' are validated on the second-highest volume exchange then I consider them 'valid'. Conversely, it would be useless to attempt market manipulation from an exchange like TradeHill as it would just get eaten up by arbitrage orders and not affect Mt. Gox at all. It is a cheap and easy way to do data filtering.
If it is durable enough to move TradeHill prices much from when I used them last then I will take another look.
Right now, MtGox is at 14 and TradeHill is at 14.2
Right now the daily model does not produce an output that is more predictive than the previous day's price.
Human emotion is a difficult thing to predict
Well, I'll tell you what then. There is one thing that has worked better than anything I have come up with for predicting the day's price, at least lately... That is, yesterday's price.
Then your models are no good..?
To my eyeballs, the pattern we see today best matches December. Even when viewed on a log scale, this summer pattern is larger and longer than any previous pattern. The former jumped ~5x within a month and plateaued for two months while around May prices jumped ~20x within two months. Does it reasonably follow that we can expect this plateau to last another month or more? Chodpaba, would that sort of thinking fit your models?
Thanks for your answers and for sharing your results...
Your models have yet to guide me toward profit, but in the last month I have lost most money in direct proportion to the my deviance from your models.
What I put in this thread about this ongoing research should not be used as guidance. ... If I thought that it could be absolutely relied upon for guidance I would keep it to myself so as not to reveal my market positions. But as it is, this is just me thinking out loud about some ongoing research... Carry on.
Your models have yet to guide me toward profit, but in the last month I have lost most money in direct proportion to the my deviance from your models.
Code:
2011-07-22 $13.463 $13.982 | ?--.--- ?--.---
2011-07-23 $13.343 $14.282 | $13.187 $14.096
2011-07-24 $13.360 $14.767 | $13.167 $14.546
2011-07-25 $13.409 $14.980 | $13.198 $14.641
2011-07-26 ?--.--- ?--.--- | $14.035 $15.487
New information set your short term sights lower. What might have set the sites higher or the same? I'm curious how well you can fudge future data. Are your inputs the full colorful range of ticks every second, or hourly or daily ranges? Can you say that IFF we have W range tomorrow, then I see X going forward, but IFF Y tomorrow, then Z is likely over the next few days?
Fingers crossed 
We are long overdue to cross into $14-$15 territory by my accounting. But that is ok... The longer we take to build up a market imbalance the more spectacular the ensuing rally will be. I am rather hoping it does not happen too soon.
I just hope it holds off until my funds make it to Dwolla.
Jump to: