Camouflaged Indicators - To improve trend indication, perhaps?

[ObseloV]

A question surrounding the non-comparison of frequency and timing of the PE market place, and RL market places. Stock markets close at set times each business day, but this concept doesn't exactly apply to us. PE has market vibrations 24/7.

I'm not 100% sure that most PE players would be bothered to follow a day-to-day movement of a specific item in the market. I wouldn't want to have one indicator dropping consistently for a week, then upping fractionally the next day, to only show the green arrow (in effect, hiding the action of the previous week)? I may have logged on a week ago, and here I see the indicator showing that the market went up. I'm also not sure that a 7 day indicator is much better, where should one be drawing the line?

Since this question is merely a matter of displaying database results, it's not farfetched to add such logic to the actual webpage, letting visitors switch to their style that they had want to view the results of. Perhaps a "basic" and an "advanced" mode will do?

In getting to a pure system, perhaps this can be automated by finding a ratio formula per game item, that indicates it's avg frequency vs avg quantity (or something).

For example, GSI and ME has a high turnover on auction, Iron a bit lower, and then at the end of the spectrum we have things like Rugaritz. Setting the ratio definition, can then possibly automatically choose an indicator range to use, based on the turnover ratio.

But perhaps we're not there yet.

[elisa]

Originally Posted by ObseloV

I'm not 100% sure that most PE players would be bothered to follow a day-to-day movement of a specific item in the market. I wouldn't want to have one indicator dropping consistently for a week, then upping fractionally the next day, to only show the green arrow (in effect, hiding the action of the previous week)? I may have logged on a week ago, and here I see the indicator showing that the market went up. I'm also not sure that a 7 day indicator is much better, where should one be drawing the line?

To avoid the confusion with continuous daily price fluctuation you may display 2 values: 1) trend (maybe not even as an arrow, but as a numeric value, because in meduim to long term you'd really like to know not only if the price goes up or down, but at what extent as well) 2) last 24-hour shift (as an arrow, as you do now I suppose).

According to my observation last 24-hour price movement is not important in PE at all (not counting maybe just some extreme events, like changes in skill system and imp guns dropping from snables), just because the market isn't big and isn't active enough to allow daily speculation. But it's getting closer My guess would be 2.5-3 days (for commodities with high turnover) - that's the period of time atm when the market reacts on the price movement.

Originally Posted by ObseloV

In getting to a pure system, perhaps this can be automated by finding a ratio formula per game item, that indicates it's avg frequency vs avg quantity (or something).

2 points about that:

1) surely it is possible to create such a set of formulae, but it will be quite subjective and I'm not sure if all the people will find it useful the same way. And there's no other way around unfortunately, as any formula would be based on the model and the model is only your own approximation of the real life (ok, in this case, virtual life ) processes.

2) if somebody provides a detailed analysis of PE market (something like Buy&Sell list for PE commodities, for instance), it should be exposed as some sort of additional service, not as a part of core system, with well explained approach. So if some players like it, they use it (in addition to, let's say, non-biased market indicators of the core system). That's how it works in some RL financial institutions: there are 1)just numbers, facts 2) models, explaining these numbers 3) smartass nerds who know everything about particular sectors of the market. But for the moment let's concentrate on the first 2 positions

[Ironheart]

I know for myself their are certain ores I want to know day to day changes on and others that I only want a much longer term view. Also, it certain that different users will have different needs, the Trader is much more likely to want the shorter timescales for example.

Would it be possible to have a 'trend period' dropdown, say 1 hr, 1 day, 5 days, 1 month etc. which would then give a trend based upon the users own preference?

Dont know if their is a serious overhead for recalculating the trend?

We did this at PERC if I'm reading you right. Calculating the trend (and thereby predicting market direction) turned out to be very expensive in terms of processor time. I suggest extensive use of hash tables to take the load of the raw data. Run a scheduled task to recreate lookup tables and show the data from there, that way you get instant loading of the resultset, even tho tha data actually takes a good few minutes to create.

[Devil Doll]

My thoughts...

IMO you should show each individual item results on a daily, weekly, monthly, quarterly and annual basis. You will need a set end time on all of these calcs to get the indicator to have much of a meaning at all. Additionally, just because an item does not sell often or does not have a price change does not mean the results are not valuable. In analyzing any market one of the factors an investor takes into account is volume. How much of a stock is moving. This too would be a valuable number to display. If there is no change in the price the data should illustrate that.

This website is a valuable resource and may give you some ideas.

http://quote.morningstar.com/Quote.h...&byrefresh=yes

Slightly off topic, but one of my main interests whenever the subject of the PE market comes up is the ability to have market indexes and stock like calculations.

I would love to see returns on a index type basis. Such as a return for the overall mineral trends, gun trends, armor trends ect. Another interesting thing that could lead to ingame stock trading would be to come up with a valuation of the companies represented in project entropia based upon their products sold. For intance, a value placed upon the Omegaton Stock. If one could demonstrate a reliable and accurate system, MA may take notice and seek a partnership.

Anyways, just some food for thought.

DD

[ObseloV]

Originally Posted by Dirk

Calculating the trend (and thereby predicting market direction) turned out to be very expensive in terms of processor time.

Our private beta draws it's charts based on three trendlines monitored for each item. The history data is updated in intervals of four hours, to be able to plot the node values needed for when a chart is requested.

For that matter, the indicator panels on the MyTwoPecs.com homepage are tracked in the same way, just getting the change in the last x number of days for the specific item. Effectively it does echo the hashtables concept as Dirk had suggested.

Right now, the plan is to keep poll history data for each item in game, with different poll type categories.

For example, when live, a record should will exist ...
- for each hour (24 per day)
- for each fourth hour (six per day)
- for each twelfth hour (two per day)
- for each day
- for each week
... in the timeline.

In each type, it would be the average price for the period that the poll record represents, not just the current price at that exact moment.

The reason this has been planned anyway, is to allow the chart factory to be able to select less nodes from the database when it attempts to draw a chart for a long period of time. Currently, every single node over a period gets drawn, which is overkill when too many nodes exist for the pixel width of the graph.

Approximately 900 nodes on a 674px wide graph...

Click to enlarge

Switching the date range of the indicator panels (since it would use the same information as the chart factory) would add the same load on the database server, but it will probably be called much more often than charts would.

So, if stress testing indicates that it is still too much processing work to get the results when the server is loaded, pre-cached panels can exist for the specific set of date ranges (1xhour, 4xhour, 12x hour, 24x hour, etc).

We'll find the solution in there somewhere. A someone once said ... There's more than one way to skin a cat.