Executive Summary
\n\n True Bearing Insights (TBI)\n and IFCHOR have partnered to provide\n market indicators for C10 and C14 Baltic routes, delivering their\n clients distinct strategic and commercial insights within the Capesize\n physical markets. TBI and IFCHOR have developed market indicators for\n each route one to two weeks forward, with models updating\n such assessments on a daily basis.\n
\n\n True Bearing Insights (TBI)\n and IFCHOR have partnered to provide\n market indicators for C10, C14 Baltic routes and the composite Capesize \n 5TC index, delivering their clients distinct strategic and commercial insights\n within the Capesize physical markets. TBI and IFCHOR have developed market indicators\n for each route and index one to two weeks forward, with models \n updating such assessments on a daily basis.\n
\n\n True Bearing Insights (TBI)\n provides market indicators for P1A, P3A Baltic routes and the \n composite Panamax 5TC index, delivering their clients distinct \n strategic and commercial insights within the Panamax physical markets. \n TBI has developed market indicators for each route and \n index one to two weeks forward, with models updating such assessments \n on a daily basis.\n
\n\n True Bearing Insights (TBI)\n provides market indicators for S2, S8 Baltic routes and the \n composite Supramax 10TC index, delivering their clients distinct \n strategic and commercial insights within the Panamax physical markets. \n TBI has developed market indicators for each route and \n index one to two weeks forward, with models updating such assessments \n on a daily basis.\n
\n\n True Bearing Insights (TBI)\n provides market indicators for the composite Capesize \n 5TC index and composite Panamax 5TC index, delivering their clients distinct \n strategic and commercial insights within the Capesize and Panamax physical markets. \n TBI has developed market indicators for each index one to two weeks forward, with models updating such assessments \n on a daily basis.\n
\nDashboard Overview
\n\n \n TBI and IFCHOR have created an interactive dashboard to display these\n market indicators, as well as additional data related to the indicator\n models themselves. \n \n \n TBI and IFCHOR have created an interactive dashboard to display these\n market indicators, as well as additional data related to the indicator\n models themselves. \n \n \n TBI has created an interactive dashboard to display these\n market indicators, as well as additional data related to the indicator\n models themselves. \n \n \n TBI has created an interactive dashboard to display these\n indicators, as well as additional data related to the indicator\n models themselves. \n \n \n TBI has created an interactive dashboard to display these \n indicators, as well as additional data related to the indicator\n models themselves. \n \n The dashboard consists of three main sections:
\n
- \n
- \n Indicator Data -\n displays data related to present and past indicators, updated daily\n \n
- \n Feature Importance Data -\n summarizes the underlying inputs driving the indicator\n \n
- \n Model Performance Data -\n details the indicator's recent performance\n \n
![\"Dashboard](\"@/assets/dashboard_display.png\"\n)
\n ![\"Dashboard](\"@/assets/dashboard_display_capesize.png\"\n)
\n \n ![\"Dashboard](\"@/assets/dashboard_display_panamax.png\"\n)
\n ![\"Dashboard](\"@/assets/dashboard_display_supramax.png\"\n)
\n \n The following provides a detailed overview of this dashboard, and\n the interpretations associated with its different elements.\n \n
Indicator Data
\n\n The main section of the dashboard - Indicator Data - contains\n current and previous indicator data within the context of underlying\n market conditions. There are two main graphics, which are accessed\n via the button on the top left of the page of the dashboard:
\n
- \n
- Present Indicator Dials \n
- Previous 3 Day Indicator Dials \n
![\"Indicator](\"@/assets/forecast_data.png\"\n)
\n ![\"Indicator](\"@/assets/forecast_data_capesize.png\"\n)
\n \n ![\"Indicator](\"@/assets/forecast_data_panamax.png\"\n)
\n ![\"Indicator](\"@/assets/forecast_data_supramax.png\"\n)
\n \n \n
- \n
- \n Indicator Ranges - indicate the indicator's range of spot prices\n within the target week\n
- \n
- \n 50% Range = dark blue shade within exterior circle; numeric\n range detailed in center of dial\n \n
- \n 70% Range = light blue shade within exterior circle\n \n
- \n The indicator's range on average should capture 50, 70% of spot prices within the target week\n \n
\n - \n Indicator Average and Direction - indicate the average\n spot price of the target week and its direction from current\n spot price.\n
- \n
- \n Average - green / red arrow between exterior and\n interior circles\n \n
- \n Direction - indicated by a) color of average arrow\n (green = up, red = down), and b) arrow direction / color in\n the center of dial\n \n
\n - \n Current Spot Price\n
- \n
- \n Black line within the interior circle; numerically detailed\n in center of dial\n \n
\n
\n
\n
![\"Indicator](\"@/assets/forecast_w1.png\"\n)
\n ![\"Indicator](\"@/assets/forecast_w1_no_label.png\"\n)
\n ![\"Indicator](\"@/assets/forecast_w2.png\"\n)
\n ![\"Indicator](\"@/assets/forecast_w2_no_label.png\"\n)
\n \n
- \n
- \n Current Indicator Range and Spot Price - exterior circle\n
- \n
- \n Current Indicator Range - dark, light blue shades (50, 70%)\n within exterior circle\n \n
- \n Current Spot Price - black line within exterior circle\n \n
\n - \n Previous Indicator Ranges and Spot Prices - three interior circles\n corresponding to previous three days\n
- \n
- \n Ordered from most recent (outside) to least recent (inside)\n \n
- \n Previous Indicator Ranges - dark, light blue shades (50, 70%)\n within interior circles\n \n
- \n Previous Spot Prices - black line within interior circles\n \n
\n - \n Indicator Trend - directional change in current indicator relative\n to three days prior\n
- \n
- \n Directional change is measured on the basis of the indicated\n average spot price; green = indicator trending up,\n red = indicator trending down\n \n
\n
\n
\n
\n Users can access additional information regarding an indicator’s\n historical behavior via the dashboard’s Date Tool. By default,\n the dashboard will always display the most recent indicators\n available, the date of which is found in the upper right header.\n Users have the ability to change the display settings to any date\n in the past (up to 2017-01-01), however, which will update the\n dashboard to reflect the data on that date. To do this, users can\n click on the \"Setting\" icon next to the present date, which will\n expand a calendar selection tool as shown below:\n
\n![\"Date](\"@/assets/calendar.png\"\n)
\n ![\"Date](\"@/assets/calendar_capesize.png\"\n)
\n \n ![\"Date](\"@/assets/calendar_panamax.png\"\n)
\n ![\"Date](\"@/assets/calendar_supramax.png\"\n)
\n \n
\n By selecting a date in the past the Present Indicator dials will\n also contain another element - the actual realized spot prices\n relative to the historical indicators. The orange dots and line\n indicate the realized spot prices and their average, respectively.\n The numeric average of spot prices is also detailed in the center\n of the dial\n
\n\n This is a visual tool to inspect behavior at discrete points in\n time, but it is important to remember that any individual day is\n one sample out of thousands, and model performance should be\n assessed over the long run in order to draw appropriate conclusions.\n
\nFeature Importance Data
\n\n The second section of the dashboard - Feature Importance Data -\n contains donut charts (one per target) highlighting the\n distribution of high-level feature categories that are most\n relevant in the indicator over time.\n
\n\n The feature categories consist of:\n
\n
- \n
- \n Target - technical / statistical properties related to the\n target's (e.g. C10, C14) price history\n \n
- \n Target - technical / statistical properties related to the\n target's (e.g. C10, C14, C5TC) price history\n \n
- \n Target - technical / statistical properties related to the\n target's (e.g. C5TC, P5TC) price history\n \n
- \n Target - technical / statistical properties related to the\n target's (e.g. P1A, P3A, P5TC) price history\n \n
- \n Target - technical / statistical properties related to the\n target's (e.g. S2, S8, S10TC) price history\n \n
- \n Baltic Indices - various composite / individual route prices\n across different segment vessels (Capesize, Panamax, etc)\n \n
- \n Baltic Futures - various composite / individual route futures\n prices across different vessel segments and time horizons\n \n
- \n Commodity Futures - raw materials, metals, energy, agriculture\n and additional commodity futures across time horizons\n \n
- \n Other - engineered / aggregated vessel data\n \n
- \n Equities / ETFs - equities and ETFs in related\n upstream industries\n \n
\n Below is an example of feature importance pie charts for C10 weeks one and two:\n
\n\n Below is an example of feature importance pie charts for weeks one and two:\n
\n\n Below is an example of feature importance pie charts for weeks one and two:\n
\n\n Below is an example of feature importance pie charts for weeks one and two:\n
\n\n Below is an example of feature importance pie charts for weeks one and two:\n
\n![\"Feature](\"@/assets/feature_importance.png\"\n)
\n ![\"Feature](\"@/assets/feature_importance_capesize.png\"\n)
\n \n ![\"Feature](\"@/assets/feature_importance_panamax.png\"\n)
\n ![\"Feature](\"@/assets/feature_importance_supramax.png\"\n)
\n \n It is important to note that within each category, hundreds of\n features are represented, including not only the raw feature\n itself (e.g. C10 spot price) but various transformations of the\n raw feature as well (e.g. technical analyses (moving average\n spreads), statistical analyses (decompositions), etc).\n
\n\n Additionally, not all features in a category will affect the\n prediction in the same direction, and in many cases features\n within a category will have different coefficient signs\n (i.e. negative vs positive correlations). Therefore, it should\n not be interpreted, for example, that if a model that relies\n heavily on baltic futures this means that baltic futures and\n indicators are heavily correlated in one direction - this is\n not the case.\n
\n\n A more appropriate interpretation of the feature importance chart,\n for example, is to use it to assess market conditions that may be\n outside the predictive model's scope. For instance, if there is\n an event unfolding in the market that falls within the categories\n defined, the indicators should be met with more assurance as the\n model has likely experienced / accounted for such relationships\n in the past, whereas events that happen outside the categories\n that are defined should be met with more caution.\n
\nModel Performance Data
\n\n The third section of the dashboard - Model Performance Data -\n contains three measures that assess different aspects of the\n predictive model's performance; these include:\n
\n
- \n
- \n Directional Accuracy - represents whether the model correctly\n indicates the average spot price per week up / down relative\n to average price of current week (i.e. is it getting direction\n right?)\n \n
- \n Indicator Accuracy - represents the error between the indicated\n average spot price per week versus the actual average spot price\n per week (i.e. is the indicator error low?)\n \n
- \n Range Accuracy - represents the alignment between the range\n targets (50 and 70%), and the proportion of actual individual\n spot prices that fell within the predicted range(s) for the\n target week (i.e. is the indicated range appropriate?)\n \n
\n The performance measures are expressed in two variations -\n relative and absolute - and rely on recent performance windows\n of either six or twelve months.\n
\n\n The relative performance measures represent the percentile rank\n of the model’s recent performance relative to its total historical\n performance (2018-01 to present). In this way, we are making the\n interpretation of each metric standardized across targets and weeks\n as follows:\n
\n
- \n
- \n 0% = model is currently performing at its worst level ever\n regarding the metric in question\n \n
- \n 50% = model is currently performing at its average level\n regarding the metric in question\n \n
- \n 100% = model is currently performing at its best level ever\n regarding the metric in question\n \n
\n The relative performance dials identify (in red) when a metric is\n below 20% (i.e. model is currently performing at the lower end of\n its usual performance), and the following \"rule of thumb\"\n interpretations should be used to assess comprehensive model\n performance:\n
\n
- \n
- \n All metrics above 20% = model behaving as expected\n \n
- \n 1 metric below 20% = model may be starting to exhibit lower\n performance, but still valid\n \n
- \n 2 metrics below 20% = model is starting to exhibit lower\n performance, use more caution in model outputs\n \n
- \n 3 metrics below 20% = model is exhibiting lower performance\n than usual, worth a deeper investigation into model behavior,\n use caution\n \n
\n The absolute performance measures represent the nominal value of\n the various metrics in order to assist in more direct\n interpretation and relate relative performance figures towards.\n In this view of model performance, the interpretation of each\n metric is as follows:\n
\n
- \n
- \n Directional Accuracy = average directional accuracy\n \n
- \n Indicator Accuracy = 1 - median absolute percentage error (MedAPE)\n \n
- \n Range Accuracy = 1 - abs(average difference between the\n target confidence interval (i.e. 50%) and percentage of actual\n values captured)\n \n
\n Therefore, similar to relative performance metrics, the absolute\n metrics have been structured so that the greater the nominal\n value the better the model performance.\n
\n\n Below is an example of how the performance dials may be represented\n in the dashboard. Each circle represents a performance metric\n corresponding to the legend on top. The largest number within the\n circle is an average of the three performance metrics, with each\n individual performance metric listed below.\n
\n![\"Model](\"@/assets/model_performance.png\"\n)
\n ![\"Model](\"@/assets/model_performance_capesize.png\"\n)
\n \n ![\"Model](\"@/assets/model_performance_panamax.png\"\n)
\n ![\"Model](\"@/assets/model_performance_supramax.png\"\n)
\n Review
\n\n The dashboard contains information related to the indicators\n for routes C14 and C10 one to two weeks forward.\n
\n\n The dashboard contains information related to the indicators\n for routes C14, C10 and the Capesize composite index one to two weeks forward.\n
\n\n The dashboard contains information related to the indicators for the Capesize composite \n index and Panamax composite index one to two weeks forward.\n
\n\n The dashboard contains information related to the indicators\n for routes P1A, P3A and the Panamax composite index one to two weeks forward.\n
\n\n The dashboard contains information related to the indicators\n for routes S2, S8, and the Supramax composite index one to two weeks forward.\n
\n\n The main commercial element of the dashboard is the indicator visual\n itself, however, this is supported with contextual model\n information related to the model's underlying features and its\n recent performance.\n
\n\n The dashboard is most useful when employed consistently as an input\n into decision processes with the understanding that while short\n term performance may fluctuate, the models deliver consistent\n forward-looking commercial insights / advantages over the long\n run - selective use of the dashboard may reduce / eliminate\n such advantages.\n
\n