AgFlow
AgFlow

Do Not Base Your Risk Management On Incomplete Forward Curves

by

Sep 4, 2020 | Commodity Trading 101

Reading time: 6 minutes

Risk analysis and risk management are core needs of anyone who wants to trade or hedge on any commodity market. For risk managers, traders, and analysts, one of the most significant market risks is failing to read a given situation correctly due to a lack of information or a lack of existing or underlying data. That is why AgFlow developed a machine learning algorithm that fills the data and information gaps in incomplete cash forward curves. Our algorithm generates new synthetic cash forward data points every regular trading day; each data point is then back-tested with real live and historical data to ensure robustness.

Let us demonstrate how it works by comparing two price series:

  • Brazil Corn FOB Brazil
  • Argentina Feed Barley FOB Argentina

We will study the reliability of each price series’ data compared to the synthetic data generated by AgFlow’s algorithm and whether they are suitable to support forward risk transactions.

 

Working Around Incomplete Forward Curves

To illustrate why AgFlow developed a machine learning-enabled risk management tool, let us observe the behavior of the Brazil Corn and Argentina Feed Barley price series’ curves in a regular two-dimensional environment.

 

Figure 1: Two-dimensional price series plots
Brazil Corn FOB Brazil
Argentina Feed Barley FOB Argentina
(end 2013 – Sep 2020) 

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ arrow_nav=”on” advanced_effect=”1″ _builder_version=”4.4.3″ custom_margin_tablet=”10px||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|tablet” global_colors_info=”{}”][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/slide-figure-1-1.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/slide-figure-1-2.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][/dica_divi_carousel]

While interesting, the historical price series in Figure 1 are not suitable for risk analysis in two respects:

  1. They do not reveal the amount of data that is missing
  2. They do not allow in-depth data granularity analysis

That is why we have chosen to represent the Brazil Corn and Argentina Barley Feed price series in a three-dimensional surface plot. Three-dimensional surface plots bring a higher granularity level to the table, highlighting where data points are missing and, therefore, enabling a more efficient, in-depth analysis. The selected window ranges from June 1 to June 10, 2020, with forward prices up to one year in the future.

Figure 2: Three-dimensional plots forward prices series
Brazil Corn FOB Brazil
Argentina Feed Barley FOB Argentina
(Jun 1 – Jun 10) 

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ arrow_nav=”on” advanced_effect=”1″ _builder_version=”4.4.3″ custom_margin=”8px||||false|false” global_colors_info=”{}”][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/BR_all.gif” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/AR_all.gif” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][/dica_divi_carousel]

The holes observed in figure 2 for the Brazil Corn and Argentina Feed Barley forward prices highlight missing data. In practice, if we were to rely solely on available data, as shown in figure 1, we would be missing a significant portion of information. The amount of missing data that was impossible to detect in the two-dimensional plots is now apparent. And as such, any hedging or trading strategy based on the incomplete series would have been precarious at best.

 

Figure 3: Representation of the amount of available forward price per day
Brazil Corn FOB Brazil
Argentina Feed Barley FOB Argentina
(Jun 20) 

 

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ arrow_nav=”on” advanced_effect=”1″ _builder_version=”4.4.3″ custom_margin=”-42px||||false|false” custom_margin_tablet=”0px||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” global_colors_info=”{}”][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/slide-3-1.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/slide-3-2.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][/dica_divi_carousel]

The amount of forward price available each day of June for each price series varies between 0 and 7 for Brazilian Corn and 0 and 12 for Argentina Feed Barley. Each commodity lacks forward price data, and at different points in time. However, they have a commonality: complete cash forward curves are not available on most trading days.

Try AgFlow’s Risk Management Solution 

✓ No credit card required

Figure 4: Availability of prices in %
Brazil Corn FOB
Argentina Feed Barley FOB

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ advanced_effect=”1″ _builder_version=”4.11.3″ custom_margin_tablet=”10px||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|tablet” hover_enabled=”0″ global_colors_info=”{}” custom_margin=”||50px||false|false” image_sizing=”79%” sticky_enabled=”0″][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2021/11/Capture-d’écran-2021-11-18-à-16.12.33.png” image_lightbox=”on” _builder_version=”4.11.3″ hover_enabled=”0″ global_colors_info=”{}” title_text=”Capture d’écran 2021-11-18 à 16.12.33″ sticky_enabled=”0″ header_text_align=”center” social_media_alignment=”center” button_alignment=”center” body_text_align=”center” subtitle_text_align=”center”][/dica_divi_carouselitem][/dica_divi_carousel]

The table represents the percentage of available prices in each series and the missing ones compared to the ideal-world number of quotations that any risk analyst or trader would have liked to work with between May and July 2020. In reality, any hedging or trading strategy based on these price series would have only captured half of the picture, making decision-making challenging at best.

Even though risk managers and traders can apply several methods to compensate for the lack of information and approximate commodities’ market value, such as:

  • Finding a related instrument
  • Marking to the last trade
  • Or even finding a physical cash assessment

However easy-to-apply, these workarounds come with significant disadvantages – they are tedious and increase the risk of introducing errors in the analysis process.

These tasks are tedious because risk managers and traders have to collect the most accurate data from every available source every day. Once the data is collected, they attempt to fill the gaps in the forward curves by interpolating and extrapolating; working with available data.

On the other hand, while these simplistic techniques provide data rapidly – give or take 20 minutes for two price series – they yield a higher standard deviation. These deviations ultimately increase the risk of producing inaccurate analysis and therefore are not suitable for managing investment risks.

This issue is the reason why at AgFlow, we apply complex models produced with machine learning algorithms.

Simple workaround techniques provide rapid but simplistic results; furthermore, they are prone to potentially yield higher errors. For these reasons, we have decided to use more complex models to obtain more precise results and mitigate errors, hence at the origin of AgFlow’s risk management creation.

Figure 5: Comparison Table Mean Absolute Error & Mean Squared Error yielded by simple techniques v. AgFlow’s risk management model

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ advanced_effect=”1″ _builder_version=”4.11.3″ custom_margin_tablet=”10px||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|tablet” hover_enabled=”0″ global_colors_info=”{}” sticky_enabled=”0″][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2021/11/Capture-d’écran-2021-11-18-à-16.27.49.png” image_lightbox=”on” _builder_version=”4.11.3″ hover_enabled=”0″ global_colors_info=”{}” title_text=”Capture d’écran 2021-11-18 à 16.27.49″ sticky_enabled=”0″][/dica_divi_carouselitem][/dica_divi_carousel]

When backtesting data produced by interpolation or extrapolation with actual cash market data, we demonstrate that workaround techniques produce data that deviate from the actual cash market prices significantly more than the data produced by AgFlow’s model. In other words, simple models yield higher errors than complex models.

However, complex models have their shortcomings too. They require more resources to run, such as inputs, time, and computational power. As such, they are manageable for a couple of price series, but they do not scale very well in-house for any stakeholder managing hundreds of price series.

Lastly, even though the error delta between simple and complex models just might not seem worth the development trouble, the rationale supporting complex models becomes hard to question when comparing synthetic data with actual cash market price data.

 

AgFlow’s Risk Management Solution

AgFlow’s solution removes all the hurdles risk managers and traders face when working with incomplete cash forward curves. As a result, they do not have to collect data from multiple sources or having to worry about computational resources and can reduce the risk of introducing errors in analyses.

AgFlow risk analysis model produces a 1-year ahead forward curve for 137 price series, every regular trading day before midday (CET). The literature-based algorithm we have developed does not just make simple calculations – as shown above – it uses multiple external inputs and historical data to build the most robust historical curves. The algorithm then feeds the curves into a machine learning model to generate a predictive forward curve tailored to tickers’ market prices. Every day, the predictions are backtested against historical data. Thanks to the backtesting, the data corrected is integrated into the model, improving it every day.

Figure 6: Two-dimensional plots comparing price actual price series with prices series generated by simple and complex models 
Brazil Corn FOB Brazil 
Argentina Feed Barley FOB

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ arrow_nav=”on” advanced_effect=”1″ _builder_version=”4.4.3″ custom_margin=”3px||-42px||false|false” custom_margin_tablet=”” custom_margin_phone=”-15px||||false|false” custom_margin_last_edited=”on|desktop” global_colors_info=”{}”][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/Figure-6.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/Figure-6-2.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][/dica_divi_carousel]

The simple and complex models’ prices series overall follow the original series very closely. However, the complex model is a stronger predictor for unusual actual data prices.

Figure 7: Three-dimensional plots representing price series using AgFlow data
Brazil Corn FOB Brazil
Argentina Feed Barley FOB Argentina

[dica_divi_carousel show_items_desktop=”1″ show_items_tablet=”1″ arrow_nav=”on” advanced_effect=”1″ _builder_version=”4.4.3″ custom_margin=”3px||-42px||false|false” custom_margin_tablet=”” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” global_colors_info=”{}”][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/Figure-5-2.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][dica_divi_carouselitem button_url_new_window=”1″ image=”https://www.agflow.com/wp-content/uploads/2020/09/Figure-7.png” image_lightbox=”on” _builder_version=”4.4.3″ global_colors_info=”{}”][/dica_divi_carouselitem][/dica_divi_carousel]

Figure 7 shows the price surface created by the AgFlow risk analysis model. Comparing the AgFlow data surfaces with actual cash market price series’ three-dimensional plots clearly illustrates that no data points are missing as the surfaces no longer contain holes. This evidence means that the machine learning model provides a price point for every regular trading day.

Try AgFlow’s Risk Management Solution 

Try AgFlow’s Risk Management Solution 

✓ No credit card required

Conclusion

When managing risk associated with cash quotes, there are two options to build complete cash forward curves:

You can either collect prices available from several sources. However, incomplete data sets are error-prone and therefore increases the risk related to transactions. Furthermore, building complete one-year cash forward curves requires significant computational resources, which can be overly time-consuming.

The other option is to use the AgFlow risk management solution. The product helps risk managers to overcome all the disadvantages of more simplistic models by providing a 1-year forward on every regular trading day curve for an ever-growing variety of price series. Thus it eliminates the time and computational resources required to build simpler models that do not come as close to real data as AgFlow’s model or forecasting price events.