Search the Community
Showing results for tags 'forecasting'.
Found 2 results
Minimum search term is 4 characters long. Can't find what you want? Click here for the custom google search instead.
This post concerns price prediction for XRP, but also -- taking a step back to provide a little perspective -- (price) prediction in general. I don't consider 'XRP to the moon!' or 'Bitcoin to USD 500,000!' to fall into the category of serious predictions. Requirements for prediction Predicting a price for XRP or any other crypto asset is a difficult task. Asking people here who may be more knowledgeable about the inner workings of Ripple (the company) or who may have been speculating in crypto for longer than oneself to make predictions for next month, next year, or several years out is something of enormous complexity that is, most likely, beyond pretty much anyone's level of skill to accomplish at this point in time. No one is immune from harboring a hope or hazarding a guess as to eventual valuations, but it is important to realize a guess is just that. Some guesses are based on better information, specific domain (banking) knowledge, and/or made from a more level-headed point of view, and thus are more educated. They are still guesses, however. Coming up with well-founded predictions hinges on at least two things: 1) availability of historical data and 2) an understanding of the relative importance of the factors involved in determining price and of their relationships. With both of these well in hand -- and if the underlying assumption described below holds -- we can take a shot at a price projection by building a predictive model. Without them, we are most likely back to 'XRP to the moon!' Obstacles to prediction What are some of the possible obstacles we may run into? As to #1, data may not be available for a long enough time period to help us discern trends, cyclic variations, seasonality, etc. It may also happen that we simply don't have enough data, we have little data that are reliable, or we completely lack trusted data sources. And what about data that may not have predictive value? We may have data with a lot of noise (extraneous factors we have a hard time detecting, or filtering even if detected) or that, coming from a variety of sources, requires extensive prepping to be usable. Data preparation, or data wrangling as it is sometimes called, may involve discarding information wholesale because parts of it are incomplete, or force us to do any number of interpolations, which detract from the data's exactitude. As to #2, not all important factors may be observable to us (ex. Ripple's possible partnership agreements in development and/or subject to NDAs), their relative importance may grow or wane over time, new/unforeseen factors may come into play (ex. SEC regulations), or we may be unable to assess how these factors interact and quantitatively influence price without designing fairly complex statistical experiments. If all of this is under control, so to speak, and we are reasonably confident in our grasp of the subject matter, we can proceed to build a model that may have a chance of formulating useful predictions. Basics of building models Predictive models are built in which current predictions account both for the additive effect of past values of variables and past predictions. These models are error-driven by the difference between previous 'wrong' predictions and previous actual values, with past data also weighed differently according to recency -- typically, more recent data are assumed to be more influential and older data are 'discounted' and incorporated into the model accordingly. Note, however, that including data from an earlier time period not resembling the current one will likely influence the model's prediction accuracy negatively. Many of these models work in real-time, meaning predictions are recalculated and the model is updated whenever new data come in (every minute, etc.) As a side note, 'real-time', commonly thought of as being on the order of milliseconds or nanoseconds, means nothing more than a time frame short enough to enable you to take a decision that can usefully act upon your system of interest. What this means is that, if things change once a day instead of every few minutes, there is no reason to measure and recalculate your model every second. As one looks increasingly ahead, the prediction horizon moves as well ('receding horizon' modeling.) In understanding and 'regressing' (finding/fitting) possible relationships between all influencing variables and the dependent/output variable (ex. price), much work is involved in gauging which among these factors may carry more weight and need to be included in the model and which can be safely ignored with a view to simplifying the model and making it less computationally intensive. Per Occam's razor, a simpler model that explains a situation 'well enough' for our purpose is preferred to a more complex one, and a good modeler must know not only what and how to model but also when diminishing returns set in and thus when to stop. A common caveat here is not to confuse correlation with causation, meaning not to jump to unwarranted conclusions about the information you think you understand and attribute non-existing properties to relationships. We should strive to see what's there, not just what we want to see that supports our biases. Human nature being what it is, this does not come easy to many. Limitations of models Many financial models of exceeding complexity have been built, with a large number being unsuccessful or only modestly successful. This is not necessarily due to lack of skill on the part of the modelers, but rather to the inherent complexity of the task, including 1) accounting properly for all important variables (many of which may be hidden/unobservable), and 2) being able to quantify/measure them accurately, both in an absolute sense as well as relative to one another. Once built, the range of validity of a model is another important aspect to consider. In our over-zealousness to draw conclusions that apply 'always', we often extrapolate conclusions beyond the realm of the model's credibility. Misapplication of models does happen, and is likely to happen more often the more a model is complex and appears as a 'black box' to the often naive end user. Simple models you can understand and explain, complex ones are beyond most people's reach. I include in the latter, for example, models based on neural nets, with 'reasoning paths' that are very hard to explain and therefore are not readily accepted in the lawsuit-prone medical diagnostics domain (vs. decision trees, which are more understandable, at least up to a point.) As someone said succinctly, you can usually torture data to the point where it will eventually confess and admit to anything you want or hope for, and it is only an understanding of the domain and modeling that can guide you here. These factors are all the more true when people are involved, and especially in an unregulated and sentiment-driven environment such as crypto, with allegiance to specific 'coins' and disdain for others bordering on fanaticism. Even stocks and bonds, which are heavily regulated and long established, pose a formidable challenge to successful prediction. Typically, mathematical models are built to handle 'steady state' situations, but cannot deal with 'transients' nearly as well or at all. Transients are sudden changes that eventually die out but reflect the fact that the system being looked at is moving from one situation or set of operating conditions to another (ex. ICO period to 5 years post-ICO.) The more stable and under control a situation is, the easier it may be to make a prediction. This, by the way, is true not only for finance but also in science and engineering. Interestingly, the more a model is good at tracking, the worse it usually is at detecting/predicting unforeseen changes. And 'unforeseen' is the key word. I always remember Peter Lynch, the successful (and very hard working) Fidelity Magellan fund manager, stating ruefully and not with little irony that 'earnings surprise' were the two words most often heard together in that setting, despite all of his and his staff's poring virtually 24x7 over financial reports, traveling extensively to talk to managers and do tire-kicking and look under hoods at company plants, etc. Consider also that these assessments and attempts at forecasting concerned established companies, within well developed and understood markets, and with years of widely available and endlessly dissected information. I should also add that there is no shortage of commercial models/tools that are retrofitted to past data and made to look as if they are near-infallible for marketing purposes, yet when applied to new data they fall abysmally short of inflated expectations. One consequence of 'over-fitting' is typically the lack of ability of a model to generalize to different situations. This is an interesting rabbit hole to explore, but I will leave it at that for now. Underlying assumptions I mentioned steady state and transients. The implicit assumption underlying prediction efforts -- and it is a huge one, ignored at one's own peril -- is that the future has a fairly close relationship to the past and present in terms of the operating environment, influencing factors, etc. I repeat, the future is assumed to be not too different from the present or the historical past. Another word for this is 'stability'. So, one may well want to ask oneself the following: How stable is the crypto environment? Is this an incremental change or a disruptive sea change we are living? Do I have abundant data, and do I know these data to be accurate and consistent across time periods? Do the data have predictive value, or is the price of any crypto asset, including XRP, a 'random walk' and therefore not useful for arriving at future prices? How confident am I that, looking at the past history of XRP prices and other information at my disposal, I can come up with something (a model or series of quantitative relationships based on the data available to me) from which I can reasonably predict what its price is likely to be 6 months from now? Is a future value prediction even a reasonable thing to attempt, or is a broad range of value the best I can hope for? Do I understand what's at play here as far as factors influencing price movement? Can I even tell how much of XRP's price movement is due to Ripple/XRP exclusively and how much to Bitcoin-specific developments? How much is there that I simply don't know? (impossible to know, by definition, but always good to ponder.) Do I understand what Ripple's value proposition is, and do I believe it will be more/less/equally interesting to FIs three years from now? How competitive is the playing field for use cases other than cross-border payments? Would I have known 2 years ago or 1 year ago where XRP's price would be today if I had applied my current reasoning to the situation then? Could well-thought-of Ripple employees have done any better with more information available to them? Can I readily quantify the potential impact on price of something quantitatively known well in advance, such as the coming escrow/lockup? If not, what are my chances of being successful with 'information' that is far more uncertain? Also, will it rain tomorrow at 10 am? Even here, weather predictions coming out of complex models are of the type '50% chance of rain' whereas we would like to know 'yes in the late morning, no in mid-afternoon', not quite the same thing. Reasonable expectations I have known CEOs and CFOs who wanted an analysis to yield 'a number' that would be an accurate prediction of some future value. They acted as if we lived in a certain world. Because we do not, it is often better to come up with an estimate of a range of values, each with a different probability of occurrence. This is because there is so much uncertainty as to what matters and how it may evolve that 'a range' is literally the best one can aim for. Claiming the ability to do otherwise would be naive and/or dishonest. Even using Monte Carlo simulation of thousands or tens of thousands of combinatorial scenarios as a means of dealing with uncertainty in relevant factors requires a level of knowledge about the data being used and the interrelationship of variables at play. Uncertainty in the inputs translates to uncertainty in the output (prediction) if you incorporate it in the model, and into a poor prediction if you don't. It is best not to ignore uncertainties, and try to identify them and act to reduce their impact instead (ex. increase the information content of the model.) Models can also be developed to be robust in the face of uncertainty, which is a complete field of research unto itself. No doubt companies such as BlackRock, GS, Fidelity, and others wading into this space will be deploying their best resources to grow the business of crypto prediction, most likely via machine learning. Massive computing resources are being and will be thrown at the task of semi-automatically building very complex predictive models. And they'll still get things wrong a fair amount of the time, despite the terabytes of data at their disposal. Regardless of the domain, it is generally easier to make predictions 'in the aggregate' about a sector or a group (of stocks, bonds, hospital patients, real estate in a specific neighborhood, and perhaps a subset of crypto assets) than it is to do so about a single individual or instance in that domain. We might be able to predict that the crypto space, as it relates to the top 50 assets, will likely 'be up' in market cap a year from now far more easily than 'by how many' dollars or cents a single asset will climb or fall in the coming week. This, of course, may disappoint those who demand or pretend to know where XRP's price will be tomorrow at 5 pm. Parting thoughts Expecting other speculators on these forums -- or Ripple's CEO, for that matter -- to come up with a number that may fairly accurately reflect XRP's value (or that of any other budding crypto project, as I am loath to call most of these initiatives 'coins') at some near future date is a fool's errand, so to speak (no offense meant.) I understand the dreams, the concerns, the impatience, and all the rest, but awareness is needed as to the complexity of what is being asked. And no one here needs to feel bad if they can't come up with a proper prediction, given all the unknowns and the fast-changing environment in which we in crypto are immersed. In my view, a sensible approach is the following: as long as our research about the company, competitors, partnerships, upcoming regulations, and so on presents convincing arguments to our reasoning and supports our expectations for a better future (qualitatively, not quantitatively), consider staying involved. When the quantitative panorama clears and our grasp of the situation improves, decide whether and how much to invest. Whenever data cease to inform, or if we are unable to establish anything with certainty, stop and reconsider. Disclaimer: I am not a financial adviser and nothing in this post, or other posts by me, constitutes financial advice. I leave you with four pithy quotes that have stood the test of time, as uttered by people well known in the world of logic, finance, the business of investing, and the area of model building/forecasting for their (un)common sense as well as their skill: On modeling: All models are wrong. Some are useful. On forecasting: If you are going to forecast, forecast often. Again, on forecasting: A forecast tells you more about the person making it than it does about the future. On overly complex explanations: When you hear hoof-beats, think horses, not zebras. Food for thought? I hope I didn't put too many here to sleep with this and that at least a few among you found it somewhat useful. It is not a complete picture, but it highlights a few aspects worth reflecting upon. Best wishes to all with Ripple and XRP.
Reuters article this morning entitled 'U.S. Treasury considering all tools to stop North Korea financing: official'. If the US, in cooperation with its global finance partners, choose to "choke" digital currencies via all means available, isn't there a fair risk that this entire marketplace gets disrupted until the threat of Mr. Kim is mitigated? - conduct DDOS on major exchanges - disallow BTC transactions - put all currencies in "transfer pergatory" - pump false alts as honey pots for the N. Korean hacks - etc. etc. Does the possibility of specific global threats impact your investments? Does the possibility of certain global threats outweigh others for you, or do you simply "index them" and go hell-bent-for-leather, as it used to be said in the US of A? Your opinions, please. "Hello, is this the NSA? Got anything, anything really strong to throw at the crypto world for a few months? Oh, I should call Israel? Or Russia? Thanks. I think I'll have M5 do that for us."